Vector engine benchmark: eshoponweb

Run started 5 Oct 2026, 10:18 UTC. Data: eShopOnWeb, 524 vectors. Queries: 20 labelled questions.

Used by: published-2026-10-08 (basis v6)

The runs of v6 were also used by the set blocked-2026-10-05-v6, which was not published; its verdict, design/verdicts/v6-verdict.md, holds the word BLOCK.

sources

file design/verdicts/v6-verdict.md#BLOCK = BLOCK
consolidated consolidated:reuse[session=v6].verdict = design/verdicts/v6-verdict.md
consolidated consolidated:reuse[session=v6].blockedFolders[0] = blocked-2026-10-05-v6

Corrections to statements recorded in this run

This run's report is kept as it was written. A set that uses this run states that the recorded statements below are false, misleading or unbacked. Each is marked in the report text.

False: Weaviate, durability. Recorded: weaviate.compose.yaml sets no persistence variable deploy/engines/weaviate.compose.yaml sets PERSISTENCE_DATA_PATH to /var/lib/weaviate. The compose file does set a persistence variable. Listed in published-2026-10-08. Marked in the report text, lines: 1.
sources
- file deploy/engines/weaviate.compose.yaml#PERSISTENCE_DATA_PATH: /var/lib/weaviate = PERSISTENCE_DATA_PATH: /var/lib/weaviate
- results results@20261005-101815-eshoponweb:targets[weaviate].durability#sets no persistence variable = sets no persistence variable
False: framing line. Recorded: Measured end to end through each engine's .NET client The statement is false for 8 of the 19 engines of this run: chroma, clickhouse, elasticsearch, milvus, opensearch, typesense, vespa and weaviate are reached through the benchmark's own HttpClient REST code, not through a .NET client of the engine. For those engines the figure is the cost of the benchmark's own request code. A code line of each is cited. Listed in published-2026-10-08. Marked in the report text, lines: 1.
sources
- file src/GenericVectorBuilder.Engines/Sinks/ChromaSink.cs#new HttpClient { BaseAddress = new Uri( options.BaseUrl = new HttpClient { BaseAddress = new Uri( options.BaseUrl
- file src/GenericVectorBuilder.Engines/Sinks/ClickHouseSink.cs#new HttpClient { BaseAddress = new Uri( $"http://{options.Host}:{options.Port}/" = new HttpClient { BaseAddress = new Uri( $"http://{options.Host}:{options.Port}/"
- file src/GenericVectorBuilder.Engines/Sinks/ElasticsearchRest.cs#new HttpClient { BaseAddress = new Uri( _baseUrl + "/" ) = new HttpClient { BaseAddress = new Uri( _baseUrl + "/" )
- file src/GenericVectorBuilder.Engines/Sinks/MilvusSink.cs#new HttpClient { BaseAddress = new Uri( options.BaseUrl = new HttpClient { BaseAddress = new Uri( options.BaseUrl
- file src/GenericVectorBuilder.Engines/Sinks/OpenSearchRest.cs#new HttpClient { BaseAddress = new Uri( _baseUrl + "/" ) = new HttpClient { BaseAddress = new Uri( _baseUrl + "/" )
- file src/GenericVectorBuilder.Engines/Sinks/TypesenseRest.cs#new HttpClient { BaseAddress = new Uri( _baseUrl + "/" ) = new HttpClient { BaseAddress = new Uri( _baseUrl + "/" )
- file src/GenericVectorBuilder.Engines/Sinks/VespaRest.cs#_data = new HttpClient( handler ) = _data = new HttpClient( handler )
- file src/GenericVectorBuilder.Engines/Sinks/WeaviateSink.cs#new HttpClient { BaseAddress = new Uri( options.BaseUrl = new HttpClient { BaseAddress = new Uri( options.BaseUrl
Not backed by a saved source: index. Recorded: an earlier run gave about 0.09 at 100,000 No log was saved for this figure, and the summary does not print the clause. A saved re-run of the MariaDB effort test (10 repeats, 524 and 2,000 random 1024-dimension vectors) read recall@10 at ef 100 of 0.988 to 0.996 at 524 vectors and 0.894 to 0.932 at 2,000. The 100,000-row figure was not re-measured the same way and is dropped. Listed in published-2026-10-08. Marked in the report text, lines: 2.
sources
- file design/bench-inputs/mariadb-effort-2026-10-07/SOURCES.txt#for which no log was saved = for which no log was saved
- file design/bench-inputs/mariadb-effort-2026-10-07/SOURCES.txt#The 100,000-row figure was not re-measured the same way, so the report drops it. = The 100,000-row figure was not re-measured the same way, so the report drops it.
- file design/bench-inputs/mariadb-effort-2026-10-07/run.log#RECALL@10 on 524 random 1024-dim vectors, 50 queries: ef 20 0.872, ef 100 0.988, ef 3200 0.988 = RECALL@10 on 524 random 1024-dim vectors, 50 queries: ef 20 0.872, ef 100 0.988, ef 3200 0.988
- file design/bench-inputs/mariadb-effort-2026-10-07/run.log#RECALL@10 on 524 random 1024-dim vectors, 50 queries: ef 20 0.886, ef 100 0.996, ef 3200 0.996 = RECALL@10 on 524 random 1024-dim vectors, 50 queries: ef 20 0.886, ef 100 0.996, ef 3200 0.996
- file design/bench-inputs/mariadb-effort-2026-10-07/run.log#RECALL@10 on 2000 random 1024-dim vectors, 50 queries: ef 20 0.468, ef 100 0.894, ef 3200 0.994 = RECALL@10 on 2000 random 1024-dim vectors, 50 queries: ef 20 0.468, ef 100 0.894, ef 3200 0.994
- file design/bench-inputs/mariadb-effort-2026-10-07/run.log#RECALL@10 on 2000 random 1024-dim vectors, 50 queries: ef 20 0.482, ef 100 0.932, ef 3200 0.992 = RECALL@10 on 2000 random 1024-dim vectors, 50 queries: ef 20 0.482, ef 100 0.932, ef 3200 0.992
- file deploy/bench/recorded-text-classes.json#"drop": "mariadb-recall" = "drop": "mariadb-recall"

Machine: CPU Intel(R) Xeon(R) CPU E5-1620 v3 @ 3.50GHz; Logical CPUs 8; RAM (GiB) 62.7; OS Ubuntu 24.04.5 LTS, kernel 6.8.0-142-generic; Governor performance; Partition client 0-1,4-5; engines 2-3,6-7; Release build; Exact mode seconds 60.

Warm-up, as this run's notes record it: Warm-up and settle check, untimed, right before every timed pass: the pass's own search at the pass's own number of searchers for at least 15 s and at least 20 searches (at most 120 s), read in windows of at least 2 s and 100 searches; then a 3 s trial of the same pass.

No CPU clock pin is recorded in this run's notes.

Engine	p50 (ms)	Searches per second, one searcher	Searches per second, eight searchers at once	Exact mode p50 (ms)	Client CPU per search, one searcher (ms)	Recall in hits	Flags
Chroma	2.02	490	808	-	0.46	200 of 200
ClickHouse	4.50	215	480	6.06	0.82	198 of 200	Un
DuckDB	3.51	282	602	4.02	4.84	200 of 200
Elasticsearch	1.31	751	2,697	1.26	0.45	200 of 200	Ix
MariaDB	0.62	1,573	5,944	1.58	0.49	200 of 200
Milvus	2.21	429	1,285	-	0.54	200 of 200
MongoDB Atlas Local	1.39	704	2,089	1.23	0.30	200 of 200
OpenSearch	1.96	497	1,539	2.63	0.45	200 of 200
Oracle 23ai Free	0.89	1,097	2,826	2.36	0.87	200 of 200	Un
Qdrant (HNSW)	0.92	1,073	3,466	0.89	0.73	200 of 200
Qdrant (exact)	0.88	1,123	3,296	-	0.74	200 of 200
Redis redis holds its data in memory: its saved docs page says 'Redis is an in-memory but persistent on disk database', and its compose file sets save "300 1" and appendonly no. sources doc `doc:design/engine-docs/redis-faq-2026-10-07.html#Redis is an in-memory but persistent on disk database` = Redis is an in-memory but persistent on disk database file `deploy/engines/redis.compose.yaml#--save "300 1"` = --save "300 1" file `deploy/engines/redis.compose.yaml#--appendonly no` = --appendonly no	0.37	2,595	5,039	0.36	0.55	200 of 200	Un
SQL Server 2025	3.89	250	613	-	1.18	200 of 200
SQL Server 2025 + DiskANN	3.62	273	690	3.58	1.01	193 of 200
Typesense	4.03	245	594	5.46	0.51	200 of 200
Vespa	1.83	527	1,595	1.84	0.92	200 of 200
Weaviate	5.19	171	312	-	0.59	200 of 200
pgvector	0.85	1,151	4,050	2.17	0.68	200 of 200
sqlite-vec	1.81	544	1,159	1.81	1.93	200 of 200

Engines are listed in alphabetical order. The table does not rank them.

A dash means the table has no figure there.

The small markers after an engine name are flags. Hover a marker for its evidence, or read the list below the tables.

Un unsettled-target The engine did not pass its own settle check before a timed pass. The evidence names the check.
Ix index-not-ready The engine did not report a finished index after the load or after the searches.

Evidence behind the flags

ClickHouse unsettled-target latency had NOT settled when timing began for default@8 (each by its own warm-up at its own concurrency and a trial of the same pass within 10% of the warm-up's settled figure). default@1: warm-up 15.0 s and 3,232 searches (0 failed) at 1 searcher; p50 of the last windows 4.436, 4.452, 4.476 ms (windows of at least 2 s and 100 searches); trial of 655 searches in 3.0 s at 1 searcher: p50 4.479 ms against the settled p50 4.452 ms, 1% apart (limit 10%); exact: warm-up 15.0 s and 2,442 searches (0 failed) at 1 searcher; p50 of the last windows 6.029, 6.056, 6.035 ms (windows of at least 2 s and 100 searches); trial of 457 searches in 3.0 s at 1 searcher: p50 6.120 ms against the settled p50 6.035 ms, 1% apart (limit 10%); default@8: warm-up 15.0 s and 7,381 searches (0 failed) at 8 searchers; QPS of the last windows 435, 510, 524 (windows of at least 2 s and 100 searches) (its last 3 windows differed by more than 5%); trial of 1,509 searches in 3.0 s at 8 searchers: 501 QPS against the settled 510 QPS, 2% apart (limit 10%), so the warm-up was EXTENDED once (32.0 s and 15,463 searches (0 failed) at 8 searchers; QPS of the last windows 440, 445, 524 (windows of at least 2 s and 100 searches), its last 3 windows differed by more than 5%); second trial of 1,583 searches in 3.0 s at 8 searchers: 526 QPS against the settled 445 QPS, 15% apart (limit 10%); still disagreeing after the extension. Its numbers may still include warm-up; rerun before quoting them.
Elasticsearch index-not-ready afterLoad: the engine's index state read not ready, 0 of 524 vectors indexed
Elasticsearch index-not-ready afterSearch: the engine's index state read not ready, 0 of 524 vectors indexed
Oracle 23ai Free unsettled-target latency had NOT settled when timing began for default@8 (each by its own warm-up at its own concurrency and a trial of the same pass within 10% of the warm-up's settled figure). exact: warm-up 15.0 s and 6,064 searches (0 failed) at 1 searcher; p50 of the last windows 2.371, 2.350, 2.331 ms (windows of at least 2 s and 100 searches); trial of 1,244 searches in 3.0 s at 1 searcher: p50 2.356 ms against the settled p50 2.350 ms, 0% apart (limit 10%); default@1: warm-up 15.0 s and 16,495 searches (0 failed) at 1 searcher; p50 of the last windows 0.888, 0.870, 0.904 ms (windows of at least 2 s and 100 searches); trial of 3,251 searches in 3.0 s at 1 searcher: p50 0.896 ms against the settled p50 0.888 ms, 1% apart (limit 10%); default@8: warm-up 15.3 s and 44,258 searches (0 failed) at 8 searchers; QPS of the last windows 3,062, 2,834, 2,622 (windows of at least 2 s and 100 searches) (its last 3 windows differed by more than 5%); trial of 9,357 searches in 3.0 s at 8 searchers: 3,118 QPS against the settled 2,834 QPS, 9% apart (limit 10%), so the warm-up was EXTENDED once (120.2 s and 344,269 searches (0 failed) at 8 searchers; QPS of the last windows 3,394, 2,474, 2,998 (windows of at least 2 s and 100 searches), the 120 s cap ran out); second trial of 9,022 searches in 3.0 s at 8 searchers: 3,006 QPS against the settled 2,998 QPS, 0% apart (limit 10%); still disagreeing after the extension. Its numbers may still include warm-up; rerun before quoting them.
Redis unsettled-target latency had NOT settled when timing began for exact (each by its own warm-up at its own concurrency and a trial of the same pass within 10% of the warm-up's settled figure). exact: warm-up 15.0 s and 41,314 searches (0 failed) at 1 searcher; p50 of the last windows 0.336, 0.374, 0.372 ms (windows of at least 2 s and 100 searches) (its last 3 windows differed by more than 5%); trial of 8,416 searches in 3.0 s at 1 searcher: p50 0.335 ms against the settled p50 0.372 ms, 11% apart (limit 10%), so the warm-up was EXTENDED once (30.0 s and 81,883 searches (0 failed) at 1 searcher; p50 of the last windows 0.363, 0.369, 0.338 ms (windows of at least 2 s and 100 searches), its last 3 windows differed by more than 5%); second trial of 8,081 searches in 3.0 s at 1 searcher: p50 0.369 ms against the settled p50 0.363 ms, 2% apart (limit 10%); still disagreeing after the extension; default@1: warm-up 15.0 s and 38,516 searches (0 failed) at 1 searcher; p50 of the last windows 0.372, 0.407, 0.378 ms (windows of at least 2 s and 100 searches) (its last 3 windows differed by more than 5%); trial of 7,842 searches in 3.0 s at 1 searcher: p50 0.377 ms against the settled p50 0.378 ms, 0% apart (limit 10%), so the warm-up was EXTENDED once (32.0 s and 82,511 searches (0 failed) at 1 searcher; p50 of the last windows 0.385, 0.380, 0.391 ms (windows of at least 2 s and 100 searches)); second trial of 7,681 searches in 3.0 s at 1 searcher: p50 0.390 ms against the settled p50 0.385 ms, 1% apart (limit 10%); settled after the extension; default@8: warm-up 15.0 s and 75,402 searches (0 failed) at 8 searchers; QPS of the last windows 4,844, 5,008, 5,010 (windows of at least 2 s and 100 searches); trial of 14,794 searches in 3.0 s at 8 searchers: 4,929 QPS against the settled 5,008 QPS, 2% apart (limit 10%). Its numbers may still include warm-up; rerun before quoting them.

Full results

This report is printed as the run wrote it, except for any sentence a note above it says was left out. This page does not check the engine texts in it. The summary page's facts table gives the label of each fact it uses.

Run 2026-10-05T10:18:15Z (run-all). Command line:

~/gvb-work/lanes/v6-final/src/GenericVectorBuilder.Bench/bin/Release/net10.0/GenericVectorBuilder.Bench.dll run-all --pipeline eshoponweb --queries golden --seed 603 --out ~/ForClaude/GenericVectorBuilder/bench-results

Machine: bench-host, Intel(R) Xeon(R) CPU E5-1620 v3 @ 3.50GHz (8 logical CPUs), 62.7 GiB RAM, GPU NVIDIA GeForce RTX 3060, 12288 MiB, Ubuntu 24.04.5 LTS, kernel 6.8.0-142-generic, .NET 10.0.12, load average at start 1.45 3.00 3.58
Data: 524 vectors x 1024 dims, collection gvbbench_eshoponweb. Read READ-ONLY from GenericVectorBuilder.dbo.gvb_eshoponweb in ChunkId order, vectors as native SqlVector<float>, 0.2 s.
Queries: golden: 20 labelled questions from ~/ForClaude/evalkit/questions_golden.json, embedded with qwen3-emb-0.6b (cached in ~/gvb-data/bench-cache/golden-eshoponweb-68df42ca69efa079.json, no embedding calls). Top 10. Throughput at concurrency 1, 8 for 20 s each.
Request speed on a small collection (524 vectors): Measured end to end through each engine's .NET client [Correction 2]; at this size it reflects per-request cost including the client library, not index scaling.
Client CPU per search is the CPU time the test's .NET client itself used for each search, measured in the same pass as the figure beside it.
Ground truth: brute force over every vector in memory, 0.0 s for all 20 queries.
nDCG@10 of the exact answer itself (the ceiling for this embedder): 0.518

engine	index	load rows/s	p50 ms	p95 ms	p99 ms	QPS@1	QPS@8	client CPU ms/search@1	client CPU ms/search@8	recall@10	nDCG@10	RAM	disk
sql	exact VECTOR_DISTANCE cosine, no vector index (full scan)	563	3.89	4.51	5.82	249.9	612.5	1.18	1.18	1.000	0.518	532.6 MiB	5.15 MiB
opensearch	faiss HNSW float32, no compression, m=16, ef_construction=128, cosinesimil; search k=top, ef_search=100; 1 shard, 0 replicas; graph built at any segment size (approximate_threshold=0); force-merged to one segment after the load	388	1.96	2.24	2.85	497.2	1539.4	0.45	0.49	1.000	0.518	2.57 GiB	9.55 MiB
duckdb	HNSW (vss extension) FLOAT[n] metric=cosine m=16 ef_construction=128, ef_search=100 per connection, persistent (hnsw_enable_experimental_persistence=true, checkpoint_threshold=256MB); exact mode = array_cosine_similarity sequential scan; score = 1 - cosine distance; searches run concurrently, one connection per searcher (opened as searchers arrive, at most 32), writes run one at a time and never overlap a search	1,546	3.51	3.87	4.96	282.1	601.9	4.84	6.61	1.000	0.518	-	2.9 MiB
mongodb	vectorSearch index, HNSW maxEdges=16 numEdgeCandidates=128, float32 binData, cosine, numCandidates=20x hits (min 100); exact mode = $vectorSearch exact:true	3,173	1.39	1.62	1.94	703.9	2088.9	0.30	0.29	1.000	0.518	632.5 MiB	0 B
chroma	HNSW M=16 ef_construction=128, ef_search=100 (Chroma default), cosine; approximate only (no exact mode)	895	2.02	2.28	2.52	489.9	807.5	0.46	0.46	1.000	0.518	54.53 MiB	414.16 KiB
mariadb	VECTOR INDEX (HNSW variant) DISTANCE=cosine, M=16 (no ef_construction setting exists), mhnsw_ef_search=100 per statement (the ef 100 most engines here use, so the search effort matches; MariaDB's own default is 20; recall@10 at ef 100 falls as the set grows (random 1024-dimension vectors, measured 2026-10-04: 0.99 at 524, 0.89 to 0.92 at 2,000; an earlier run gave about 0.09 at 100,000 [Correction 3])), mhnsw_max_cache_size 4G; exact mode = IGNORE INDEX full scan	1,039	0.62	0.71	0.88	1573.4	5944.2	0.49	0.38	1.000	0.518	170.6 MiB	22.01 MiB
qdrant-hnsw	HNSW m=16 ef_construct=100, hnsw_ef=server default, cosine; indexing_threshold_kb 1 and full_scan_threshold_kb 10 (server defaults are 10,000 each) so a small collection builds and walks its graph	5,186	0.92	1.03	1.46	1073.1	3465.9	0.73	0.52	1.000	0.518	38.2 MiB	164.1 MiB
typesense	HNSW float32 (hnswlib), m=16, ef_construction=128, cosine; search k=top, ef=100; exact mode = filter ordinal:>=0 with flat_search_cutoff; index held in memory	1,043	4.03	4.36	4.87	245.0	593.7	0.51	0.57	1.000	0.518	118.4 MiB	0 B
clickhouse	vector_similarity HNSW cosineDistance, quantization bf16, M=16 ef_construction=128, hnsw_candidate_list_size_for_search=256, rescoring off; exact mode = full scan with skip indexes off	3,361	4.50	6.26	6.90	214.7	479.8	0.82	1.02	0.990	0.515	227.7 MiB	3.27 MiB
qdrant	exact scan: the builder's sink sends exact=true on every search, so no HNSW graph is used whether or not Qdrant has built one (see the index state)	9,365	0.88	0.99	1.40	1122.7	3295.5	0.74	0.53	1.000	0.518	35.72 MiB	196.08 MiB
pgvector	HNSW vector_cosine_ops m=16 ef_construction=128, hnsw.ef_search=100 per query, float32 vector(n), cosine; exact mode = same query with index scans off (sequential scan)	612	0.85	1.00	1.28	1150.9	4050.0	0.68	0.54	1.000	0.518	100.1 MiB	7.52 MiB
vespa	HNSW float32 tensor, prenormalized-angular (cosine), max-links-per-node=16, neighbors-to-explore-at-insert=128; search targetHits=top, ef=100 via exploreAdditionalHits; exact mode = approximate:false; vectors held in memory	322	1.83	2.31	2.66	527.0	1595.2	0.92	0.86	1.000	0.518	2.79 GiB	12.11 MiB
sqlitevec	vec0 brute-force scan, no ANN index (exact), float32, cosine distance, default chunk_size=1024; score = 1 - cosine distance; searches run concurrently, one WAL reader connection per searcher (opened as searchers arrive, at most 32), writes run one at a time and may overlap searches	4,041	1.81	2.04	2.26	543.7	1158.9	1.93	3.45	1.000	0.518	-	9.72 MiB
sql-diskann	DiskANN (preview) via VECTOR_SEARCH, cosine, build {"StartId":"306", "L":"48", "M":"8", "R":"48"}; the exact mode scans the same table	630	3.62	4.08	5.09	273.4	690.4	1.01	1.07	0.965	0.526	533 MiB	4.52 MiB
milvus	HNSW M=16 efConstruction=128, ef=100, metric COSINE, Strong consistency searches; approximate only (no exact mode)	1,198	2.21	2.79	4.24	428.8	1285.1	0.54	0.63	1.000	0.518	215.5 MiB	10.85 MiB
elasticsearch	HNSW float32, no quantization, m=16, ef_construction=128, cosine; search k=top, num_candidates=100; 1 shard, 0 replicas; force-merged to one segment after the load (at 1,024 dimensions a segment under 1,043 vectors gets no graph)	545	1.31	1.48	1.76	750.6	2697.5	0.45	0.49	1.000	0.518	2.58 GiB	2.39 MiB
redis	HNSW TYPE FLOAT32 M=16 EF_CONSTRUCTION=128, EF_RUNTIME=100 per query, cosine; exact mode = FLAT index built on first exact query	11,939	0.37	0.48	0.71	2594.8	5039.1	0.55	0.37	1.000	0.518	21.19 MiB	2.43 MiB
weaviate	HNSW maxConnections(M)=16 efConstruction=128, ef=-1 (dynamic: limit x 8 clamped 100..500), cosine, no quantization; approximate only (no exact mode)	1,018	5.19	10.4	11.4	171.2	312.1	0.59	0.60	1.000	0.518	69.82 MiB	2.96 MiB
oracle	HNSW in-memory neighbor graph NEIGHBORS=16 EFCONSTRUCTION=128, EFSEARCH=100 per query, cosine; exact mode = FETCH EXACT FIRST (full scan); Oracle Free caps itself at 2 CPUs (cpu_count 2 in V$PARAMETER, edition FREE in V$INSTANCE, 8 host CPUs in V$OSSTAT NUM_CPUS; the 2 CPU thread limit is Oracle's documented Free edition limit)	2,020	0.89	1.07	1.86	1096.8	2826.3	0.87	0.65	1.000	0.518	2.14 GiB	0 B

Details per target

SQL Server 2025 sql

Engine: Microsoft SQL Server 2025 (RTM-CU9) (KB5122048) - 17.0.5005.3 (X64), Enterprise Developer Edition (64-bit), in container gvb-mssql (image mcr.microsoft.com/mssql/server:2025-CU9-ubuntu-24.04@sha256:2b5b581621126574f3d1f75e78d3eebe8d05aedb59ad0cfdf9aa42cb0634d726; cpuset 2-3,6-7 from its creation; SQL Server counts 4 CPU(s) and runs 4 visible scheduler(s), affinity AUTO) (compose)
Index: exact VECTOR_DISTANCE cosine, no vector index (full scan)
Search settings: description=exact VECTOR_DISTANCE cosine, no vector index (full scan)
Load: 524 rows in batches of 1000, 0.9 s of upserts (563 rows/s); count matched 0.0 s after the last upsert; index step n/a
Search: 4999 latency samples, target held 524 rows, 0 errors
Client CPU per search: 1 searcher 1.18 ms, 8 searchers 1.18 ms
RAM: 532.6 MiB (docker stats); disk: 5.15 MiB (table and its indexes, reserved pages)
Started by run-all for this measurement with every container created on CPUs 2-3,6-7 (started mssql.compose.yaml with every container created on CPUs 2-3,6-7: gvb-mssql cpuset 2-3,6-7 (read back from docker inspect)); stopped afterwards.
Index after the load: ready, 0 of 524 indexed (no index used, exact scan by design: dbo.gvb_gvbbench_eshoponweb in GvbBench holds 524 rows; sys.vector_indexes lists no index on the table; indexes: PK_gvb_gvbbench_eshoponweb CLUSTERED, IX_gvb_gvbbench_eshoponweb_DocKey NONCLUSTERED; a real search under SET STATISTICS XML ON ran: Clustered Index Scan of GvbBench.gvb_gvbbench_eshoponweb through PK_gvb_gvbbench_eshoponweb (IndexKind Clustered), 524 rows read, no vector index operator, 10 hits). Durability: A commit returns after its transaction-log records are written to disk (SQL Server write-ahead logging; delayed durability is DISABLED); a database created here copies the model database: recovery model FULL, page_verify CHECKSUM; no global trace flags are enabled; mssql.conf of container gvb-mssql (/var/opt/mssql/mssql.conf, on the host at ~/gvb-data/engines/mssql/mssql.conf) does not exist, so it sets: nothing; it has no [control] or [traceflag] entry, so SQL Server's own Linux defaults for flushing writes apply (Microsoft's Linux performance guide names trace flag 3982 as that default; read from the guide, not tested here). Not tested by cutting power; whether the disk's own write cache reaches the media was not checked. Container settings from its environment (names only): MSSQL_AGENT_ENABLED, MSSQL_MEMORY_LIMIT_MB, MSSQL_PID, MSSQL_RPC_PORT, MSSQL_SA_PASSWORD..
Load average 1.54 2.97 3.56 (1/5/15 min) when searching began.
Rehearsal before any timed pass, untimed, every pass type at its own concurrency for at least 30 s through the same code the timed passes use: default@1 7,706 searches in 30.0 s, default@8 17,308 searches in 30.0 s.
Settle: settled before every timed pass, each by its own warm-up at its own concurrency and a trial of the same pass within 10% of the warm-up's settled figure. default@1: warm-up 15.0 s and 3,772 searches (0 failed) at 1 searcher; p50 of the last windows 3.901, 3.897, 3.855 ms (windows of at least 2 s and 100 searches); trial of 755 searches in 3.0 s at 1 searcher: p50 3.890 ms against the settled p50 3.897 ms, 0% apart (limit 10%); default@8: warm-up 15.0 s and 8,843 searches (0 failed) at 8 searchers; QPS of the last windows 586, 622, 602 (windows of at least 2 s and 100 searches) (its last 3 windows differed by more than 5%); trial of 1,855 searches in 3.0 s at 8 searchers: 616 QPS against the settled 602 QPS, 2% apart (limit 10%), so the warm-up was EXTENDED once (30.0 s and 17,163 searches (0 failed) at 8 searchers; QPS of the last windows 580, 565, 580 (windows of at least 2 s and 100 searches)); second trial of 1,806 searches in 3.0 s at 8 searchers: 599 QPS against the settled 580 QPS, 3% apart (limit 10%); settled after the extension.
Pass default@1 after rehearsal: 2026-10-05T10:19:45.215Z to 2026-10-05T10:20:05.216Z (20.0 s), 4,999 searches, 0 failed, p50 3.893 ms, mean 3.997 ms, p99 5.816 ms, 249.9 QPS (1000/QPS 4.001 ms).
Pass default@8 after default@1: 2026-10-05T10:21:11.324Z to 2026-10-05T10:21:31.333Z (20.0 s), 12,256 searches, 0 failed, p50 12.364 ms, mean 13.055 ms, p99 25.640 ms, 612.5 QPS (1000/QPS 1.633 ms).
Passes in the order run: default@1, default@8; each after its own warm-up. Untimed searches (rehearsals, per-pass warm-ups, settle checks and extensions): 68,364 sent, 0 failed (warmupErrors). Index after the searches: ready, 0 of 524 indexed (no index used, exact scan by design: dbo.gvb_gvbbench_eshoponweb in GvbBench holds 524 rows; sys.vector_indexes lists no index on the table; indexes: PK_gvb_gvbbench_eshoponweb CLUSTERED, IX_gvb_gvbbench_eshoponweb_DocKey NONCLUSTERED; a real search under SET STATISTICS XML ON ran: Clustered Index Scan of GvbBench.gvb_gvbbench_eshoponweb through PK_gvb_gvbbench_eshoponweb (IndexKind Clustered), 524 rows read, no vector index operator, 10 hits).
Benchmark copy gvbbench_eshoponweb dropped afterwards.
Connection: container address on its Docker network, not the published port (no docker-proxy): container gvb-mssql (ed0a0d981ab4) on network gvb-mssql_default at container-ip:1433, not through docker-proxy localhost:14330. Open after the passes: container-ip:1433 x9 (this target).
CPU pinning: cpuset from the container's creation when run-all started it on the engine CPUs, else docker update --cpuset-cpus (each container's change line says which), CPUs 2-3,6-7. Changed: container gvb-mssql: already on CPUs 2-3,6-7 (its cpuset since creation), not changed. Read back: container gvb-mssql: 85 thread(s) on CPUs 2-3,6-7 (every thread of the container).
Clock per pass (median MHz of engine CPUs / client CPUs, governor, CPUs busy outside the benchmark, client CPU per search): default@1 3504/3494 MHz, performance, outside load 0.16 before the warm-up, 0.16 during, client CPU 1.181 ms per search; default@8 3492/3492 MHz, performance, outside load 0.16 before the warm-up, 0.15 during, client CPU 1.177 ms per search.

OpenSearch opensearch

Engine: OpenSearch 3.9.0 (k-NN plugin, faiss) (compose)
Index: faiss HNSW float32, no compression, m=16, ef_construction=128, cosinesimil; search k=top, ef_search=100; 1 shard, 0 replicas; graph built at any segment size (approximate_threshold=0); force-merged to one segment after the load
Search settings: approximate_threshold=0, ef_construction=128, ef_search=100, k=top, m=16
Load: 524 rows in batches of 1000, 1.4 s of upserts (388 rows/s); count matched 0.1 s after the last upsert; index step 0.9 s (force-merged to one segment, graph ready after 0.9 s: every segment searched through its HNSW graph, covering 524 of 524 vectors: 1 segment(s), 0 merge(s) running, profiled probe search: ann_search_count 1 (segments tried through a graph), exact_search_count 0 (of those, scanned instead), approximate_threshold 0; force-merged to one segment on purpose, so one graph answers every search)
Search: 9945 latency samples, target held 524 rows, 0 errors
Client CPU per search: 1 searcher 0.45 ms, 8 searchers 0.49 ms
Exact mode: 22,336 searches, p50 2.63 ms, p95 2.97 ms, recall 1.000
RAM: 2.57 GiB (docker stats); disk: 9.55 MiB added by this load (11.95 MiB (whole engine data folder), 2.4 MiB before)
Started by run-all for this measurement with every container created on CPUs 2-3,6-7 (started opensearch.compose.yaml with every container created on CPUs 2-3,6-7: gvb-opensearch cpuset 2-3,6-7 (read back from docker inspect)); stopped afterwards.
Index after the load: ready, 524 of 524 indexed (every segment searched through its HNSW graph, covering 524 of 524 vectors: 1 segment(s), 0 merge(s) running, profiled probe search: ann_search_count 1 (segments tried through a graph), exact_search_count 0 (of those, scanned instead), approximate_threshold 0; force-merged to one segment on purpose, so one graph answers every search). Durability: Every acknowledged bulk request is fsynced to the translog before the answer: index.translog.durability=request, the OpenSearch default, which neither this sink nor opensearch.compose.yaml overrides (OpenSearchReadinessTests reads it back from the live index). By that setting a process crash or power loss loses no acknowledged write; this is read from the setting, not shown by pulling power. One node and no replicas, so a lost disk loses the data..
Load average 3.49 3.28 3.55 (1/5/15 min) when searching began.
Rehearsal before any timed pass, untimed, every pass type at its own concurrency for at least 30 s through the same code the timed passes use: default@8 26,897 searches in 30.0 s, exact 11,241 searches in 30.0 s, default@1 14,984 searches in 30.0 s.
Settle: settled before every timed pass, each by its own warm-up at its own concurrency and a trial of the same pass within 10% of the warm-up's settled figure. default@8: warm-up 15.0 s and 21,888 searches (0 failed) at 8 searchers; QPS of the last windows 1,546, 1,534, 1,542 (windows of at least 2 s and 100 searches); trial of 4,595 searches in 3.0 s at 8 searchers: 1,530 QPS against the settled 1,542 QPS, 1% apart (limit 10%); exact: warm-up 15.0 s and 5,597 searches (0 failed) at 1 searcher; p50 of the last windows 2.643, 2.632, 2.634 ms (windows of at least 2 s and 100 searches); trial of 1,118 searches in 3.0 s at 1 searcher: p50 2.639 ms against the settled p50 2.634 ms, 0% apart (limit 10%); default@1: warm-up 15.0 s and 7,470 searches (0 failed) at 1 searcher; p50 of the last windows 1.957, 1.947, 1.961 ms (windows of at least 2 s and 100 searches); trial of 1,503 searches in 3.0 s at 1 searcher: p50 1.965 ms against the settled p50 1.957 ms, 0% apart (limit 10%).
Pass default@8 after rehearsal: 2026-10-05T10:23:55.711Z to 2026-10-05T10:24:15.714Z (20.0 s), 30,793 searches, 0 failed, p50 5.008 ms, mean 5.194 ms, p99 9.971 ms, 1539.4 QPS (1000/QPS 0.650 ms).
Pass exact after default@8: 2026-10-05T10:24:33.728Z to 2026-10-05T10:25:33.730Z (60.0 s), 22,336 searches, 0 failed, p50 2.635 ms, mean 2.684 ms, p99 3.618 ms, 372.3 QPS (1000/QPS 2.686 ms).
Pass default@1 after exact: 2026-10-05T10:25:51.757Z to 2026-10-05T10:26:11.758Z (20.0 s), 9,945 searches, 0 failed, p50 1.964 ms, mean 2.009 ms, p99 2.855 ms, 497.2 QPS (1000/QPS 2.011 ms).
Passes in the order run: default@8, exact, default@1; each after its own warm-up. Untimed searches (rehearsals, per-pass warm-ups, settle checks and extensions): 95,293 sent, 0 failed (warmupErrors). Index after the searches: ready, 524 of 524 indexed (every segment searched through its HNSW graph, covering 524 of 524 vectors: 1 segment(s), 0 merge(s) running, profiled probe search: ann_search_count 1 (segments tried through a graph), exact_search_count 0 (of those, scanned instead), approximate_threshold 0; force-merged to one segment on purpose, so one graph answers every search).
Benchmark copy gvbbench_eshoponweb dropped afterwards.
Connection: container address on its Docker network, not the published port (no docker-proxy): container gvb-opensearch (b063d45b32bc) on network engines_default at container-ip:9200, not through docker-proxy localhost:9201. Open after the passes: container-ip:9200 x1 (this target).
CPU pinning: cpuset from the container's creation when run-all started it on the engine CPUs, else docker update --cpuset-cpus (each container's change line says which), CPUs 2-3,6-7. Changed: container gvb-opensearch: already on CPUs 2-3,6-7 (its cpuset since creation), not changed. Read back: container gvb-opensearch: 56 thread(s) on CPUs 2-3,6-7 (every thread of the container).
Clock per pass (median MHz of engine CPUs / client CPUs, governor, CPUs busy outside the benchmark, client CPU per search): default@8 3492/3492 MHz, performance, outside load 0.15 before the warm-up, 0.13 during, client CPU 0.495 ms per search; exact 3492/3492 MHz, performance, outside load 0.14 before the warm-up, 0.15 during, client CPU 0.453 ms per search; default@1 3492/3492 MHz, performance, outside load 0.15 before the warm-up, 0.16 during, client CPU 0.451 ms per search.

DuckDB duckdb

Engine: DuckDB 1.5.6 + vss b833341 (HNSW, embedded, in process) (embedded)
Index: HNSW (vss extension) FLOAT[n] metric=cosine m=16 ef_construction=128, ef_search=100 per connection, persistent (hnsw_enable_experimental_persistence=true, checkpoint_threshold=256MB); exact mode = array_cosine_similarity sequential scan; score = 1 - cosine distance; searches run concurrently, one connection per searcher (opened as searchers arrive, at most 32), writes run one at a time and never overlap a search
Search settings: checkpoint_threshold=256MB, ef_construction=128, ef_search=100, hnsw_enable_experimental_persistence=true, m=16, metric=cosine
Load: 524 rows in batches of 1000, 0.3 s of upserts (1,546 rows/s); count matched 0.0 s after the last upsert; index step 0.0 s (HNSW is maintained inside every transaction, nothing to build; confirmed in 0.01 s: duckdb_indexes() lists gvb_gvbbench_eshoponweb_hnsw, pragma_hnsw_index_info() counts 524 vectors of 524 rows, EXPLAIN of the default search shows HNSW_INDEX_SCAN on it = True)
Search: 5643 latency samples, target held 524 rows, 0 errors
Client CPU per search: 1 searcher 4.84 ms, 8 searchers 6.61 ms
Exact mode: 14,651 searches, p50 4.02 ms, p95 4.74 ms, recall 1.000
RAM: in this process, not measured; disk: 2.9 MiB added by this load (413.83 MiB (engine data folder), 410.93 MiB before)
Index after the load: ready, 524 of 524 indexed (duckdb_indexes() lists gvb_gvbbench_eshoponweb_hnsw, pragma_hnsw_index_info() counts 524 vectors of 524 rows, EXPLAIN of the default search shows HNSW_INDEX_SCAN on it = True). Durability: DuckDB's write-ahead log is fsynced at every commit and no DuckDbSink setting changes that (checkpoint_threshold=256MB only spaces out the checkpoints that write the database file; measured with strace: 42 WAL fsyncs for 40 single-record commits plus 2 setup statements), so an operating-system crash or power cut loses no committed row; measured with kill -9 right after the last commit and a reopen: all 50, 300 and 2,000 rows and an HNSW index that counted the same number were recovered from the log; not tested and documented by DuckDB: the HNSW index is file-backed only through hnsw_enable_experimental_persistence = true, which this sink turns on, and WAL recovery for such custom indexes is not complete, so a crash during a checkpoint or a later commit can damage the index while the rows survive.
Load average 1.66 3.06 3.46 (1/5/15 min) when searching began.
Rehearsal before any timed pass, untimed, every pass type at its own concurrency for at least 30 s through the same code the timed passes use: default@1 8,628 searches in 30.0 s, default@8 18,086 searches in 30.0 s, exact 7,331 searches in 30.0 s.
Settle: settled before every timed pass, each by its own warm-up at its own concurrency and a trial of the same pass within 10% of the warm-up's settled figure. default@1: warm-up 15.0 s and 4,201 searches (0 failed) at 1 searcher; p50 of the last windows 3.536, 3.516, 3.556 ms (windows of at least 2 s and 100 searches); trial of 844 searches in 3.0 s at 1 searcher: p50 3.517 ms against the settled p50 3.536 ms, 1% apart (limit 10%); default@8: warm-up 15.0 s and 8,989 searches (0 failed) at 8 searchers; QPS of the last windows 605, 600, 602 (windows of at least 2 s and 100 searches); trial of 1,805 searches in 3.0 s at 8 searchers: 600 QPS against the settled 602 QPS, 0% apart (limit 10%); exact: warm-up 15.0 s and 3,646 searches (0 failed) at 1 searcher; p50 of the last windows 4.012, 4.084, 4.035 ms (windows of at least 2 s and 100 searches); trial of 728 searches in 3.0 s at 1 searcher: p50 4.035 ms against the settled p50 4.035 ms, 0% apart (limit 10%).
Pass default@1 after rehearsal: 2026-10-05T10:28:02.271Z to 2026-10-05T10:28:22.274Z (20.0 s), 5,643 searches, 0 failed, p50 3.514 ms, mean 3.542 ms, p99 4.959 ms, 282.1 QPS (1000/QPS 3.545 ms).
Pass default@8 after default@1: 2026-10-05T10:28:40.293Z to 2026-10-05T10:29:00.300Z (20.0 s), 12,041 searches, 0 failed, p50 12.637 ms, mean 13.282 ms, p99 27.635 ms, 601.9 QPS (1000/QPS 1.661 ms).
Pass exact after default@8: 2026-10-05T10:29:18.306Z to 2026-10-05T10:30:18.308Z (60.0 s), 14,651 searches, 0 failed, p50 4.023 ms, mean 4.092 ms, p99 5.603 ms, 244.2 QPS (1000/QPS 4.095 ms).
Passes in the order run: default@1, default@8, exact; each after its own warm-up. Untimed searches (rehearsals, per-pass warm-ups, settle checks and extensions): 54,258 sent, 0 failed (warmupErrors). Index after the searches: ready, 524 of 524 indexed (duckdb_indexes() lists gvb_gvbbench_eshoponweb_hnsw, pragma_hnsw_index_info() counts 524 vectors of 524 rows, EXPLAIN of the default search shows HNSW_INDEX_SCAN on it = True).
Benchmark copy gvbbench_eshoponweb dropped afterwards.
Connection: in this process (embedded), no network: no network address. No open connection seen (no open TCP connection was found).
CPU pinning: none: an embedded engine runs inside the benchmark client process, so it shares the client CPUs, CPUs 0-1,4-5.
Clock per pass (median MHz of engine CPUs / client CPUs, governor, CPUs busy outside the benchmark, client CPU per search): default@1 3592/3592 MHz, performance, outside load 0.16 before the warm-up, 0.12 during, client CPU 4.843 ms per search; default@8 3593/3592 MHz, performance, outside load 0.15 before the warm-up, 0.16 during, client CPU 6.609 ms per search; exact 3593/3592 MHz, performance, outside load 0.14 before the warm-up, 0.15 during, client CPU 6.018 ms per search.

MongoDB Atlas Local mongodb

Engine: MongoDB 8.0.32 Atlas Local (mongod + mongot Vector Search) (compose)
Index: vectorSearch index, HNSW maxEdges=16 numEdgeCandidates=128, float32 binData, cosine, numCandidates=20x hits (min 100); exact mode = $vectorSearch exact:true
Search settings: maxEdges=16, numCandidates=20x, numEdgeCandidates=128
Load: 524 rows in batches of 1000, 0.2 s of upserts (3,173 rows/s); count matched 1.2 s after the last upsert; index step 10.3 s (index status READY, queryable True; mongot holds 524 of 524 documents in 2 segment(s) (264 + 260 documents), 2 searched through the HNSW graph (Approximate); segment layout unchanged for 10.2 s over 11 readings; wait took 10.3 s)
Search: 14079 latency samples, target held 524 rows, 0 errors
Client CPU per search: 1 searcher 0.30 ms, 8 searchers 0.29 ms
Exact mode: 47,886 searches, p50 1.23 ms, p95 1.42 ms, recall 1.000
RAM: 632.5 MiB (docker stats); disk: 0 B added by this load (1.36 GiB (whole engine data folder), 1.36 GiB before)
Started by run-all for this measurement with every container created on CPUs 2-3,6-7 (started mongodb.compose.yaml with every container created on CPUs 2-3,6-7: gvb-mongodb cpuset 2-3,6-7 (read back from docker inspect)); stopped afterwards.
Index after the load: ready, 524 of 524 indexed (index status READY, queryable True; mongot holds 524 of 524 documents in 2 segment(s) (264 + 260 documents), 2 searched through the HNSW graph (Approximate)). Durability: Acknowledged writes are journaled before the acknowledgement. The sink sets no write concern, so the server default applies: getDefaultRWConcern gives w majority and the one-member replica set (--replSet gvbmongo in the image) has writeConcernMajorityJournalDefault true, with WiredTiger journaling on (journalCommitInterval 100 ms is only the interval for unacknowledged work). Measured 2026-10-04: 30 acknowledged single-document inserts raised WiredTiger 'log sync operations' by 30 and strace showed fdatasync on /data/db/journal/WiredTigerLog files. A crash loses no acknowledged write. mongot's search index is not part of that promise: it follows the collection asynchronously and is rebuilt from it..
Load average 2.73 3.63 3.66 (1/5/15 min) when searching began.
Rehearsal before any timed pass, untimed, every pass type at its own concurrency for at least 30 s through the same code the timed passes use: default@1 17,997 searches in 30.0 s, exact 23,489 searches in 30.0 s, default@8 61,035 searches in 30.0 s.
Settle: settled before every timed pass, each by its own warm-up at its own concurrency and a trial of the same pass within 10% of the warm-up's settled figure. default@1: warm-up 15.0 s and 10,556 searches (0 failed) at 1 searcher; p50 of the last windows 1.394, 1.394, 1.387 ms (windows of at least 2 s and 100 searches); trial of 2,079 searches in 3.0 s at 1 searcher: p50 1.407 ms against the settled p50 1.394 ms, 1% apart (limit 10%); exact: warm-up 15.0 s and 11,990 searches (0 failed) at 1 searcher; p50 of the last windows 1.223, 1.226, 1.221 ms (windows of at least 2 s and 100 searches); trial of 2,387 searches in 3.0 s at 1 searcher: p50 1.234 ms against the settled p50 1.223 ms, 1% apart (limit 10%); default@8: warm-up 15.0 s and 31,067 searches (0 failed) at 8 searchers; QPS of the last windows 2,004, 2,118, 2,059 (windows of at least 2 s and 100 searches) (its last 3 windows differed by more than 5%); trial of 6,288 searches in 3.0 s at 8 searchers: 2,093 QPS against the settled 2,059 QPS, 2% apart (limit 10%), so the warm-up was EXTENDED once (30.0 s and 62,778 searches (0 failed) at 8 searchers; QPS of the last windows 2,075, 2,107, 2,092 (windows of at least 2 s and 100 searches)); second trial of 6,323 searches in 3.0 s at 8 searchers: 2,106 QPS against the settled 2,092 QPS, 1% apart (limit 10%); settled after the extension.
Pass default@1 after rehearsal: 2026-10-05T10:32:25.493Z to 2026-10-05T10:32:45.494Z (20.0 s), 14,079 searches, 0 failed, p50 1.394 ms, mean 1.418 ms, p99 1.937 ms, 703.9 QPS (1000/QPS 1.421 ms).
Pass exact after default@1: 2026-10-05T10:33:03.511Z to 2026-10-05T10:34:03.511Z (60.0 s), 47,886 searches, 0 failed, p50 1.228 ms, mean 1.251 ms, p99 1.787 ms, 798.1 QPS (1000/QPS 1.253 ms).
Pass default@8 after exact: 2026-10-05T10:35:09.575Z to 2026-10-05T10:35:29.578Z (20.0 s), 41,782 searches, 0 failed, p50 3.697 ms, mean 3.827 ms, p99 6.993 ms, 2088.9 QPS (1000/QPS 0.479 ms).
Passes in the order run: default@1, exact, default@8; each after its own warm-up. Untimed searches (rehearsals, per-pass warm-ups, settle checks and extensions): 267,311 sent, 0 failed (warmupErrors). Index after the searches: ready, 524 of 524 indexed (index status READY, queryable True; mongot holds 524 of 524 documents in 2 segment(s) (264 + 260 documents), 2 searched through the HNSW graph (Approximate)).
Benchmark copy gvbbench_eshoponweb dropped afterwards.
Connection: container address on its Docker network, not the published port (no docker-proxy): container gvb-mongodb (7aa6f7272040) on network gvb-mongodb_default at container-ip:27017, not through docker-proxy localhost:27017. Open after the passes: container-ip:27017 x10 (this target).
CPU pinning: cpuset from the container's creation when run-all started it on the engine CPUs, else docker update --cpuset-cpus (each container's change line says which), CPUs 2-3,6-7. Changed: container gvb-mongodb: already on CPUs 2-3,6-7 (its cpuset since creation), not changed. Read back: container gvb-mongodb: 192 thread(s) on CPUs 2-3,6-7 (every thread of the container).
Clock per pass (median MHz of engine CPUs / client CPUs, governor, CPUs busy outside the benchmark, client CPU per search): default@1 3492/3492 MHz, performance, outside load 0.20 before the warm-up, 0.23 during, client CPU 0.303 ms per search; exact 3492/3492 MHz, performance, outside load 0.21 before the warm-up, 0.25 during, client CPU 0.298 ms per search; default@8 3492/3492 MHz, performance, outside load 0.17 before the warm-up, 0.16 during, client CPU 0.286 ms per search.

Chroma chroma

Engine: Chroma 1.4.4 (single node, REST API v2) (compose)
Index: HNSW M=16 ef_construction=128, ef_search=100 (Chroma default), cosine; approximate only (no exact mode)
Search settings: M=16, ef_construction=128, ef_search=100
Load: 524 rows in batches of 1000, 0.6 s of upserts (895 rows/s); count matched 0.0 s after the last upsert; index step 0.0 s (Chroma has no index build to wait for and exposes no index status; checked in 0.0 s: Chroma exposes no index status (indexing_status endpoint: HTTP 500 {"error":"InternalError","message":"Method scout_logs is not implemented"}), so indexed vectors are not reported and ready only means the two counts agree to within 1 percent: count endpoint 524, a vector search asking for 524 results returned 524)
Search: 9799 latency samples, target held 524 rows, 0 errors
Client CPU per search: 1 searcher 0.46 ms, 8 searchers 0.46 ms
RAM: 54.53 MiB (docker stats); disk: 414.16 KiB added by this load (2.01 GiB (whole engine data folder), 2.01 GiB before)
Started by run-all for this measurement with every container created on CPUs 2-3,6-7 (started chroma.compose.yaml with every container created on CPUs 2-3,6-7: gvb-chroma cpuset 2-3,6-7 (read back from docker inspect)); stopped afterwards.
Index after the load: ready, ? of 524 indexed (Chroma exposes no index status (indexing_status endpoint: HTTP 500 {"error":"InternalError","message":"Method scout_logs is not implemented"}), so indexed vectors are not reported and ready only means the two counts agree to within 1 percent: count endpoint 524, a vector search asking for 524 results returned 524). Durability: SQLite rollback journal with its default synchronous=FULL under the data directory (chroma.compose.yaml sets IS_PERSISTENT=1 and PERSIST_DIRECTORY=/data and no sync setting): every write is committed to chroma.sqlite3 with fsync of the journal, the directory and the database file before the call returns (strace: 15 database fsyncs and 38 journal fsyncs for one create, four 500-vector upserts and one delete), and the HNSW files are written every sync_threshold=1000 vectors and rebuilt from the SQLite log after a crash; measured with kill -9 and a restart: all 50, 300 and 2,000 vectors were still stored and searchable.
Load average 3.47 3.86 3.73 (1/5/15 min) when searching began.
Rehearsal before any timed pass, untimed, every pass type at its own concurrency for at least 30 s through the same code the timed passes use: default@8 24,244 searches in 30.0 s, default@1 14,710 searches in 30.0 s.
Settle: settled before every timed pass, each by its own warm-up at its own concurrency and a trial of the same pass within 10% of the warm-up's settled figure. default@8: warm-up 15.0 s and 12,114 searches (0 failed) at 8 searchers; QPS of the last windows 808, 808, 806 (windows of at least 2 s and 100 searches); trial of 2,420 searches in 3.0 s at 8 searchers: 805 QPS against the settled 808 QPS, 0% apart (limit 10%); default@1: warm-up 15.0 s and 7,347 searches (0 failed) at 1 searcher; p50 of the last windows 2.008, 2.020, 2.027 ms (windows of at least 2 s and 100 searches); trial of 1,474 searches in 3.0 s at 1 searcher: p50 2.019 ms against the settled p50 2.020 ms, 0% apart (limit 10%).
Pass default@8 after rehearsal: 2026-10-05T10:37:56.174Z to 2026-10-05T10:38:16.182Z (20.0 s), 16,157 searches, 0 failed, p50 9.827 ms, mean 9.903 ms, p99 11.820 ms, 807.5 QPS (1000/QPS 1.238 ms).
Pass default@1 after default@8: 2026-10-05T10:38:34.189Z to 2026-10-05T10:38:54.190Z (20.0 s), 9,799 searches, 0 failed, p50 2.017 ms, mean 2.039 ms, p99 2.521 ms, 489.9 QPS (1000/QPS 2.041 ms).
Passes in the order run: default@8, default@1; each after its own warm-up. Untimed searches (rehearsals, per-pass warm-ups, settle checks and extensions): 62,309 sent, 0 failed (warmupErrors). Index after the searches: ready, ? of 524 indexed (Chroma exposes no index status (indexing_status endpoint: HTTP 500 {"error":"InternalError","message":"Method scout_logs is not implemented"}), so indexed vectors are not reported and ready only means the two counts agree to within 1 percent: count endpoint 524, a vector search asking for 524 results returned 524).
Benchmark copy gvbbench_eshoponweb dropped afterwards.
Connection: container address on its Docker network, not the published port (no docker-proxy): container gvb-chroma (e8999f253062) on network gvb-chroma_default at container-ip:8000, not through docker-proxy localhost:8000. Open after the passes: container-ip:8000 x8 (this target).
CPU pinning: cpuset from the container's creation when run-all started it on the engine CPUs, else docker update --cpuset-cpus (each container's change line says which), CPUs 2-3,6-7. Changed: container gvb-chroma: already on CPUs 2-3,6-7 (its cpuset since creation), not changed. Read back: container gvb-chroma: 9 thread(s) on CPUs 2-3,6-7 (every thread of the container).
Clock per pass (median MHz of engine CPUs / client CPUs, governor, CPUs busy outside the benchmark, client CPU per search): default@8 3592/3592 MHz, performance, outside load 0.08 before the warm-up, 0.08 during, client CPU 0.456 ms per search; default@1 3592/3592 MHz, performance, outside load 0.09 before the warm-up, 0.09 during, client CPU 0.456 ms per search.

MariaDB mariadb

Engine: MariaDB 11.8.9 (InnoDB + VECTOR INDEX) (compose)
Index: VECTOR INDEX (HNSW variant) DISTANCE=cosine, M=16 (no ef_construction setting exists), mhnsw_ef_search=100 per statement (the ef 100 most engines here use, so the search effort matches; MariaDB's own default is 20; recall@10 at ef 100 falls as the set grows (random 1024-dimension vectors, measured 2026-10-04: 0.99 at 524, 0.89 to 0.92 at 2,000; an earlier run gave about 0.09 at 100,000 [Correction 3])), mhnsw_max_cache_size 4G; exact mode = IGNORE INDEX full scan
Search settings: DISTANCE=cosine, M=16, mhnsw_ef_search=100
Load: 524 rows in batches of 1000, 0.5 s of upserts (1,039 rows/s); count matched 0.0 s after the last upsert; index step 0.5 s (VECTOR INDEX is maintained inside every INSERT, nothing to build; confirmed after 2 poll(s) in 0.5 s (poll 1 of 2 said: not ready: a search at mhnsw_ef_search 100 returned only 389 of 524 requested rows through the index (EXPLAIN key 'vec_idx') (information_schema lists vec_idx as INDEX_TYPE VECTOR, EXPLAIN of the default search uses key vec_idx, the server applied mhnsw_ef_search 100 (asked for 100, read back from the server), a search at that effort returned 389 of 524 requested rows through the index; MariaDB has no per-index row count (the graph is a hidden InnoDB table), so indexed vectors are not reported)): information_schema lists vec_idx as INDEX_TYPE VECTOR, EXPLAIN of the default search uses key vec_idx, the server applied mhnsw_ef_search 100 (asked for 100, read back from the server), a search at that effort returned 518 of 524 requested rows through the index; MariaDB has no per-index row count (the graph is a hidden InnoDB table), so indexed vectors are not reported)
Search: 31469 latency samples, target held 524 rows, 0 errors
Client CPU per search: 1 searcher 0.49 ms, 8 searchers 0.38 ms
Exact mode: 37,338 searches, p50 1.58 ms, p95 1.69 ms, recall 1.000
RAM: 170.6 MiB (docker stats); disk: 22.01 MiB added by this load (182.79 MiB (whole engine data folder), 160.78 MiB before)
Started by run-all for this measurement with every container created on CPUs 2-3,6-7 (started mariadb-bench.compose.yaml with every container created on CPUs 2-3,6-7: gvbbench-mariadb cpuset 2-3,6-7 (read back from docker inspect)); stopped afterwards.
Index after the load: ready, ? of 524 indexed (information_schema lists vec_idx as INDEX_TYPE VECTOR, EXPLAIN of the default search uses key vec_idx, the server applied mhnsw_ef_search 100 (asked for 100, read back from the server), a search at that effort returned 518 of 524 requested rows through the index; MariaDB has no per-index row count (the graph is a hidden InnoDB table), so indexed vectors are not reported). Durability: innodb_flush_log_at_trx_commit=2 (set in mariadb-bench.compose.yaml for the benchmark's own container gvbbench-mariadb, and in mariadb.compose.yaml for the daily gvb-mariadb, which has the same server settings): the InnoDB redo log is written to the operating system at every commit but fsynced about once a second, so a crash of the mariadbd process loses nothing, while an operating-system crash or power cut can lose the last second of commits; innodb_doublewrite is on (default), the binary log is off, and the vector graph is an InnoDB table under the same log (settings read from the running server with SHOW VARIABLES; the crash behaviour is InnoDB's documented behaviour for this setting and was not tested on either container).
Load average 2.55 3.38 3.57 (1/5/15 min) when searching began.
Rehearsal before any timed pass, untimed, every pass type at its own concurrency for at least 30 s through the same code the timed passes use: default@1 47,069 searches in 30.0 s, default@8 179,245 searches in 30.0 s, exact 18,673 searches in 30.0 s.
Settle: settled before every timed pass, each by its own warm-up at its own concurrency and a trial of the same pass within 10% of the warm-up's settled figure. default@1: warm-up 15.0 s and 23,507 searches (0 failed) at 1 searcher; p50 of the last windows 0.625, 0.627, 0.626 ms (windows of at least 2 s and 100 searches); trial of 4,749 searches in 3.0 s at 1 searcher: p50 0.624 ms against the settled p50 0.626 ms, 0% apart (limit 10%); default@8: warm-up 15.0 s and 88,833 searches (0 failed) at 8 searchers; QPS of the last windows 5,892, 5,908, 5,958 (windows of at least 2 s and 100 searches); trial of 17,909 searches in 3.0 s at 8 searchers: 5,967 QPS against the settled 5,908 QPS, 1% apart (limit 10%); exact: warm-up 15.0 s and 9,314 searches (0 failed) at 1 searcher; p50 of the last windows 1.583, 1.607, 1.572 ms (windows of at least 2 s and 100 searches); trial of 1,867 searches in 3.0 s at 1 searcher: p50 1.582 ms against the settled p50 1.583 ms, 0% apart (limit 10%).
Pass default@1 after rehearsal: 2026-10-05T10:40:51.379Z to 2026-10-05T10:41:11.380Z (20.0 s), 31,469 searches, 0 failed, p50 0.624 ms, mean 0.634 ms, p99 0.877 ms, 1573.4 QPS (1000/QPS 0.636 ms).
Pass default@8 after default@1: 2026-10-05T10:41:29.416Z to 2026-10-05T10:41:49.418Z (20.0 s), 118,892 searches, 0 failed, p50 1.300 ms, mean 1.344 ms, p99 2.465 ms, 5944.2 QPS (1000/QPS 0.168 ms).
Pass exact after default@8: 2026-10-05T10:42:07.469Z to 2026-10-05T10:43:07.470Z (60.0 s), 37,338 searches, 0 failed, p50 1.579 ms, mean 1.605 ms, p99 2.168 ms, 622.3 QPS (1000/QPS 1.607 ms).
Passes in the order run: default@1, default@8, exact; each after its own warm-up. Untimed searches (rehearsals, per-pass warm-ups, settle checks and extensions): 391,166 sent, 0 failed (warmupErrors). Index after the searches: ready, ? of 524 indexed (information_schema lists vec_idx as INDEX_TYPE VECTOR, EXPLAIN of the default search uses key vec_idx, the server applied mhnsw_ef_search 100 (asked for 100, read back from the server), a search at that effort returned 518 of 524 requested rows through the index; MariaDB has no per-index row count (the graph is a hidden InnoDB table), so indexed vectors are not reported).
Benchmark copy gvbbench_eshoponweb dropped afterwards.
Connection: container address on its Docker network, not the published port (no docker-proxy): container gvbbench-mariadb (24791ced5ae1) on network gvbbench-mariadb_default at container-ip:3306, not through docker-proxy localhost:13306. Open after the passes: container-ip:3306 x8 (this target).
CPU pinning: cpuset from the container's creation when run-all started it on the engine CPUs, else docker update --cpuset-cpus (each container's change line says which), CPUs 2-3,6-7. Changed: container gvbbench-mariadb: already on CPUs 2-3,6-7 (its cpuset since creation), not changed. Read back: container gvbbench-mariadb: 12 thread(s) on CPUs 2-3,6-7 (every thread of the container).
Clock per pass (median MHz of engine CPUs / client CPUs, governor, CPUs busy outside the benchmark, client CPU per search): default@1 3492/3492 MHz, performance, outside load 0.11 before the warm-up, 0.15 during, client CPU 0.487 ms per search; default@8 3492/3492 MHz, performance, outside load 0.14 before the warm-up, 0.08 during, client CPU 0.378 ms per search; exact 3592/3592 MHz, performance, outside load 0.11 before the warm-up, 0.13 during, client CPU 0.494 ms per search.

Qdrant (HNSW) qdrant-hnsw

Engine: Qdrant 1.17.0 in container gvb-qdrant (image qdrant/qdrant:v1.17.0@sha256:f1c7272cdac52b38c1a0e89313922d940ba50afd90d593a1605dbbc214e66ffb; cpuset 2-3,6-7 from its creation; 3 search thread(s)) (compose)
Index: HNSW m=16 ef_construct=100, hnsw_ef=server default, cosine; indexing_threshold_kb 1 and full_scan_threshold_kb 10 (server defaults are 10,000 each) so a small collection builds and walks its graph
Search settings: ef_construct=100, hnsw_ef=server, m=16
Load: 524 rows in batches of 1000, 0.1 s of upserts (5,186 rows/s); count matched 0.0 s after the last upsert; index step 2.0 s (indexing_threshold_kb 1, full_scan_threshold_kb 10; status Green, 524 of 524 vectors in HNSW segments (2 segments), waited 2.0 s)
Search: 21462 latency samples, target held 524 rows, 0 errors
Client CPU per search: 1 searcher 0.73 ms, 8 searchers 0.52 ms
Exact mode: 66,394 searches, p50 0.89 ms, p95 1.00 ms, recall 1.000
RAM: 38.2 MiB (docker stats); disk: 164.1 MiB (collection folder)
Started by run-all for this measurement with every container created on CPUs 2-3,6-7 (started qdrant.compose.yaml with every container created on CPUs 2-3,6-7: gvb-qdrant cpuset 2-3,6-7 (read back from docker inspect)); stopped afterwards.
Index after the load: ready, 524 of 524 indexed (status Green, optimizer ok, indexed_vectors_count 524 of 524 points, 2 segments, indexing_threshold_kb 1, full_scan_threshold_kb 10; every one of 1 non-empty segments has a complete HNSW graph above its full-scan threshold; no search has run yet, so the walk is proven by the segment settings only; searches counted since the collection was created: unfiltered_hnsw 0, unfiltered_plain 0, unfiltered_exact 0). Durability: What was measured (strace -f on the Qdrant server process, 30 single-point REST upserts per setting, every sync call timed against the request that caused it): with wait=true each upsert had exactly one msync(MS_SYNC) of the write-ahead-log segment inside the request, before the reply (30 of 30); with wait=false the 30 upserts were acknowledged within 0.26 s with no sync call, and the first WAL msync (one call covering all 30 records) came 2.7 s after the last reply, followed by the segment-file flushes. So the log is flushed to disk under both settings, but with wait=false the flush comes after the acknowledgement: an operating-system crash or power cut in that gap loses acknowledged writes, while a crash of the Qdrant process alone should not, because the bytes are already in the kernel's page cache (inferred, not tested). Every upsert here is sent with wait=true in batches of 256 over gRPC; the strace used one point per request, so one flush per batch is inferred, not measured. Segment files are flushed every 5 s; log segments 32 MB, 0 created ahead. /qdrant/config/config.yaml in container gvb-qdrant (the image's own file; its storage keys are listed) sets: storage.collection.quantization = null, storage.collection.replication_factor = 1, storage.collection.vectors.on_disk = null, storage.collection.write_consistency_factor = 1, storage.hnsw_index.ef_construct = 100, storage.hnsw_index.full_scan_threshold_kb = 10000, storage.hnsw_index.m = 16, storage.hnsw_index.max_indexing_threads = 0, storage.hnsw_index.on_disk = false, storage.hnsw_index.payload_m = null, storage.max_collections = null, storage.node_type = Normal, storage.on_disk_payload = true, storage.optimizers.default_segment_number = 0, storage.optimizers.deleted_threshold = 0.2, storage.optimizers.flush_interval_sec = 5, storage.optimizers.indexing_threshold_kb = 10000, storage.optimizers.max_optimization_threads = null, storage.optimizers.max_segment_size_kb = null, storage.optimizers.vacuum_min_vector_number = 1000, storage.performance.max_search_threads = 0, storage.performance.optimizer_cpu_budget = 0, storage.performance.update_rate_limit = null, storage.shard_transfer_method = null, storage.snapshots_config.snapshots_storage = local, storage.snapshots_path = ./snapshots, storage.storage_path = ./storage, storage.temp_path = null, storage.update_concurrency = null, storage.wal.wal_capacity_mb = 32, storage.wal.wal_segments_ahead = 0; no QDRANT__ environment overrides in the server process. Not tested by cutting power; whether the disk's own write cache reaches the media was not checked..
Load average 2.31 3.42 3.58 (1/5/15 min) when searching began.
Rehearsal before any timed pass, untimed, every pass type at its own concurrency for at least 30 s through the same code the timed passes use: exact 33,341 searches in 30.0 s, default@1 32,871 searches in 30.0 s, default@8 103,493 searches in 30.0 s.
Settle: settled before every timed pass, each by its own warm-up at its own concurrency and a trial of the same pass within 10% of the warm-up's settled figure. exact: warm-up 15.0 s and 16,516 searches (0 failed) at 1 searcher; p50 of the last windows 0.895, 0.899, 0.896 ms (windows of at least 2 s and 100 searches); trial of 3,302 searches in 3.0 s at 1 searcher: p50 0.897 ms against the settled p50 0.896 ms, 0% apart (limit 10%); default@1: warm-up 15.0 s and 16,100 searches (0 failed) at 1 searcher; p50 of the last windows 0.922, 0.920, 0.924 ms (windows of at least 2 s and 100 searches); trial of 3,234 searches in 3.0 s at 1 searcher: p50 0.917 ms against the settled p50 0.922 ms, 1% apart (limit 10%); default@8: warm-up 15.0 s and 51,878 searches (0 failed) at 8 searchers; QPS of the last windows 3,502, 3,456, 3,450 (windows of at least 2 s and 100 searches); trial of 10,334 searches in 3.0 s at 8 searchers: 3,442 QPS against the settled 3,456 QPS, 0% apart (limit 10%).
Pass exact after rehearsal: 2026-10-05T10:45:05.869Z to 2026-10-05T10:46:05.870Z (60.0 s), 66,394 searches, 0 failed, p50 0.891 ms, mean 0.902 ms, p99 1.444 ms, 1106.6 QPS (1000/QPS 0.904 ms).
Pass default@1 after exact: 2026-10-05T10:46:23.944Z to 2026-10-05T10:46:43.944Z (20.0 s), 21,462 searches, 0 failed, p50 0.920 ms, mean 0.930 ms, p99 1.461 ms, 1073.1 QPS (1000/QPS 0.932 ms).
Pass default@8 after default@1: 2026-10-05T10:47:01.972Z to 2026-10-05T10:47:21.973Z (20.0 s), 69,322 searches, 0 failed, p50 2.224 ms, mean 2.306 ms, p99 4.025 ms, 3465.9 QPS (1000/QPS 0.289 ms).
Passes in the order run: exact, default@1, default@8; each after its own warm-up. Untimed searches (rehearsals, per-pass warm-ups, settle checks and extensions): 271,069 sent, 0 failed (warmupErrors). Index after the searches: ready, 524 of 524 indexed (status Green, optimizer ok, indexed_vectors_count 524 of 524 points, 2 segments, indexing_threshold_kb 1, full_scan_threshold_kb 10; every one of 1 non-empty segments has a complete HNSW graph above its full-scan threshold; 308,694 segment searches walked the graph and none scanned; searches counted since the collection was created: unfiltered_hnsw 308,694, unfiltered_plain 0, unfiltered_exact 119,553).
Benchmark copy gvbbench_eshoponweb dropped afterwards.
Connection: container address on its Docker network, not the published port (no docker-proxy): container gvb-qdrant (c0db7f9c2205) on network gvb-qdrant_default at container-ip:6334, not through docker-proxy localhost:16334; container gvb-qdrant (c0db7f9c2205) on network gvb-qdrant_default at container-ip:6333, not through docker-proxy localhost:16333. Open after the passes: container-ip:6334 x1 (this target).
CPU pinning: cpuset from the container's creation when run-all started it on the engine CPUs, else docker update --cpuset-cpus (each container's change line says which), CPUs 2-3,6-7. Changed: container gvb-qdrant: already on CPUs 2-3,6-7 (its cpuset since creation), not changed. Read back: container gvb-qdrant: 24 thread(s) on CPUs 2-3,6-7 (every thread of the container).
Clock per pass (median MHz of engine CPUs / client CPUs, governor, CPUs busy outside the benchmark, client CPU per search): exact 3492/3492 MHz, performance, outside load 0.13 before the warm-up, 0.12 during, client CPU 0.726 ms per search; default@1 3492/3492 MHz, performance, outside load 0.12 before the warm-up, 0.09 during, client CPU 0.733 ms per search; default@8 3492/3492 MHz, performance, outside load 0.11 before the warm-up, 0.14 during, client CPU 0.521 ms per search.

Typesense typesense

Engine: Typesense 30.2 (vector search) (compose)
Index: HNSW float32 (hnswlib), m=16, ef_construction=128, cosine; search k=top, ef=100; exact mode = filter ordinal:>=0 with flat_search_cutoff; index held in memory
Search settings: ef=100, ef_construction=128, k=top, m=16
Load: 524 rows in batches of 1000, 0.5 s of upserts (1,043 rows/s); count matched 0.0 s after the last upsert; index step 0.0 s (nothing to build, Typesense inserts into the HNSW graph while writing; ready after 0.0 s: HNSW graph returns 524 of 524 vectors, no writes queued: num_documents 524, pending_write_batches 0, health ok, unfiltered vector query with k above the count returned 524; the graph is built in memory during each write, so there is no later build step)
Search: 4900 latency samples, target held 524 rows, 0 errors
Client CPU per search: 1 searcher 0.51 ms, 8 searchers 0.57 ms
Exact mode: 10,859 searches, p50 5.46 ms, p95 5.87 ms, recall 1.000
RAM: 118.4 MiB (docker stats); disk: 0 B added by this load (242.64 MiB (whole engine data folder), 257.6 MiB before)
Started by run-all for this measurement with every container created on CPUs 2-3,6-7 (started typesense.compose.yaml with every container created on CPUs 2-3,6-7: gvb-typesense cpuset 2-3,6-7 (read back from docker inspect)); stopped afterwards.
Index after the load: ready, 524 of 524 indexed (HNSW graph returns 524 of 524 vectors, no writes queued: num_documents 524, pending_write_batches 0, health ok, unfiltered vector query with k above the count returned 524; the graph is built in memory during each write, so there is no later build step). Durability: Every acknowledged write is appended to Typesense's raft log and fsynced before the answer: braft raft_sync=true with raft_sync_policy=0 (sync immediately), read from the running container's brpc /flags page on 2026-10-04. A restart replays the log from the last snapshot (typesense.compose.yaml sets TYPESENSE_SNAPSHOT_INTERVAL_SECONDS=300) and rebuilds the in-memory HNSW graph, so by those settings a crash loses no acknowledged write, but the restart is slow after a big load. This is read from the settings, not shown by pulling power. One node and no replicas, so a lost disk loses the data..
Load average 5.58 3.72 3.61 (1/5/15 min) when searching began.
Rehearsal before any timed pass, untimed, every pass type at its own concurrency for at least 30 s through the same code the timed passes use: default@8 17,795 searches in 30.0 s, default@1 7,329 searches in 30.0 s, exact 5,421 searches in 30.0 s.
Settle: settled before every timed pass, each by its own warm-up at its own concurrency and a trial of the same pass within 10% of the warm-up's settled figure. default@8: warm-up 15.0 s and 8,878 searches (0 failed) at 8 searchers; QPS of the last windows 590, 592, 592 (windows of at least 2 s and 100 searches); trial of 1,783 searches in 3.0 s at 8 searchers: 593 QPS against the settled 592 QPS, 0% apart (limit 10%); default@1: warm-up 15.0 s and 3,656 searches (0 failed) at 1 searcher; p50 of the last windows 4.030, 4.024, 4.033 ms (windows of at least 2 s and 100 searches); trial of 738 searches in 3.0 s at 1 searcher: p50 4.030 ms against the settled p50 4.030 ms, 0% apart (limit 10%); exact: warm-up 15.0 s and 2,718 searches (0 failed) at 1 searcher; p50 of the last windows 5.448, 5.448, 5.465 ms (windows of at least 2 s and 100 searches); trial of 547 searches in 3.0 s at 1 searcher: p50 5.461 ms against the settled p50 5.448 ms, 0% apart (limit 10%).
Pass default@8 after rehearsal: 2026-10-05T10:49:21.548Z to 2026-10-05T10:49:41.554Z (20.0 s), 11,877 searches, 0 failed, p50 13.009 ms, mean 13.472 ms, p99 25.017 ms, 593.7 QPS (1000/QPS 1.684 ms).
Pass default@1 after default@8: 2026-10-05T10:49:59.563Z to 2026-10-05T10:50:19.563Z (20.0 s), 4,900 searches, 0 failed, p50 4.028 ms, mean 4.080 ms, p99 4.874 ms, 245.0 QPS (1000/QPS 4.082 ms).
Pass exact after default@1: 2026-10-05T10:50:37.576Z to 2026-10-05T10:51:37.579Z (60.0 s), 10,859 searches, 0 failed, p50 5.460 ms, mean 5.524 ms, p99 6.727 ms, 181.0 QPS (1000/QPS 5.526 ms).
Passes in the order run: default@8, default@1, exact; each after its own warm-up. Untimed searches (rehearsals, per-pass warm-ups, settle checks and extensions): 48,865 sent, 0 failed (warmupErrors). Index after the searches: ready, 524 of 524 indexed (HNSW graph returns 524 of 524 vectors, no writes queued: num_documents 524, pending_write_batches 0, health ok, unfiltered vector query with k above the count returned 524; the graph is built in memory during each write, so there is no later build step).
Benchmark copy gvbbench_eshoponweb dropped afterwards.
Connection: container address on its Docker network, not the published port (no docker-proxy): container gvb-typesense (044be3c03b4d) on network engines_default at container-ip:8108, not through docker-proxy localhost:8108. Open after the passes: container-ip:8108 x1 (this target).
CPU pinning: cpuset from the container's creation when run-all started it on the engine CPUs, else docker update --cpuset-cpus (each container's change line says which), CPUs 2-3,6-7. Changed: container gvb-typesense: already on CPUs 2-3,6-7 (its cpuset since creation), not changed. Read back: container gvb-typesense: 303 thread(s) on CPUs 2-3,6-7 (every thread of the container).
Clock per pass (median MHz of engine CPUs / client CPUs, governor, CPUs busy outside the benchmark, client CPU per search): default@8 3592/3592 MHz, performance, outside load 0.15 before the warm-up, -0.02 during, client CPU 0.569 ms per search; default@1 3592/3592 MHz, performance, outside load 0.04 before the warm-up, 0.15 during, client CPU 0.507 ms per search; exact 3592/3592 MHz, performance, outside load 0.09 before the warm-up, 0.16 during, client CPU 0.519 ms per search.

ClickHouse clickhouse

Engine: ClickHouse 26.3.39.7 (MergeTree + vector_similarity index) (compose)
Index: vector_similarity HNSW cosineDistance, quantization bf16, M=16 ef_construction=128, hnsw_candidate_list_size_for_search=256, rescoring off; exact mode = full scan with skip indexes off
Search settings: M=16, ef_construction=128, hnsw_candidate_list_size_for_search=256
Load: 524 rows in batches of 1000, 0.2 s of upserts (3,361 rows/s); count matched 0.0 s after the last upsert; index step 6.2 s (1 data part(s), index files on 1 of them, 524 of 524 live rows in indexed parts; 0 merge(s) running, 0 unfinished mutation(s); plan of the default search uses the Skip index vec_idx on 1 of 1 parts; wait took 6.2 s)
Search: 4294 latency samples, target held 524 rows, 0 errors
Client CPU per search: 1 searcher 0.82 ms, 8 searchers 1.02 ms
Exact mode: 9,683 searches, p50 6.06 ms, p95 6.76 ms, recall 1.000
RAM: 227.7 MiB (docker stats); disk: 3.27 MiB added by this load (3.38 MiB (whole engine data folder), 112.39 KiB before)
Started by run-all for this measurement with every container created on CPUs 2-3,6-7 (started clickhouse.compose.yaml with every container created on CPUs 2-3,6-7: gvb-clickhouse cpuset 2-3,6-7 (read back from docker inspect)); stopped afterwards.
Index after the load: ready, 524 of 524 indexed (1 data part(s), index files on 1 of them, 524 of 524 live rows in indexed parts; 0 merge(s) running, 0 unfinished mutation(s); plan of the default search uses the Skip index vec_idx on 1 of 1 parts). Durability: Acknowledged inserts are not fsynced. The sink creates plain MergeTree tables, so the MergeTree settings are the defaults: fsync_after_insert 0 and fsync_part_directory 0 (system.merge_tree_settings), and async_insert 1 with wait_for_async_insert 1 is the server default for the gvb login (system.settings), so an insert is acknowledged once its part is written, not once it is on disk. Measured 2026-10-04 with strace over 30 acknowledged single-row inserts (30 Ok rows in system.asynchronous_insert_log): zero fsync, fdatasync or sync_file_range calls, while the control, a table created with fsync_after_insert=1, made 61 fdatasync calls for 5 inserts. A host power loss or kernel crash can lose acknowledged rows that are still in the OS page cache; a ClickHouse process crash alone should not (inferred, not tested)..
Load average 1.65 3.11 3.44 (1/5/15 min) when searching began.
Rehearsal before any timed pass, untimed, every pass type at its own concurrency for at least 30 s through the same code the timed passes use: default@1 6,509 searches in 30.0 s, exact 4,853 searches in 30.0 s, default@8 14,481 searches in 30.0 s.
WARNING: latency had NOT settled when timing began for default@8 (each by its own warm-up at its own concurrency and a trial of the same pass within 10% of the warm-up's settled figure). default@1: warm-up 15.0 s and 3,232 searches (0 failed) at 1 searcher; p50 of the last windows 4.436, 4.452, 4.476 ms (windows of at least 2 s and 100 searches); trial of 655 searches in 3.0 s at 1 searcher: p50 4.479 ms against the settled p50 4.452 ms, 1% apart (limit 10%); exact: warm-up 15.0 s and 2,442 searches (0 failed) at 1 searcher; p50 of the last windows 6.029, 6.056, 6.035 ms (windows of at least 2 s and 100 searches); trial of 457 searches in 3.0 s at 1 searcher: p50 6.120 ms against the settled p50 6.035 ms, 1% apart (limit 10%); default@8: warm-up 15.0 s and 7,381 searches (0 failed) at 8 searchers; QPS of the last windows 435, 510, 524 (windows of at least 2 s and 100 searches) (its last 3 windows differed by more than 5%); trial of 1,509 searches in 3.0 s at 8 searchers: 501 QPS against the settled 510 QPS, 2% apart (limit 10%), so the warm-up was EXTENDED once (32.0 s and 15,463 searches (0 failed) at 8 searchers; QPS of the last windows 440, 445, 524 (windows of at least 2 s and 100 searches), its last 3 windows differed by more than 5%); second trial of 1,583 searches in 3.0 s at 8 searchers: 526 QPS against the settled 445 QPS, 15% apart (limit 10%); still disagreeing after the extension. Its numbers may still include warm-up; rerun before quoting them.
Pass default@1 after rehearsal: 2026-10-05T10:53:41.376Z to 2026-10-05T10:54:01.378Z (20.0 s), 4,294 searches, 0 failed, p50 4.498 ms, mean 4.655 ms, p99 6.897 ms, 214.7 QPS (1000/QPS 4.658 ms).
Pass exact after default@1: 2026-10-05T10:54:19.393Z to 2026-10-05T10:55:19.398Z (60.0 s), 9,683 searches, 0 failed, p50 6.056 ms, mean 6.193 ms, p99 8.937 ms, 161.4 QPS (1000/QPS 6.197 ms).
Pass default@8 after exact: 2026-10-05T10:56:27.495Z to 2026-10-05T10:56:47.501Z (20.0 s), 9,600 searches, 0 failed, p50 16.187 ms, mean 16.666 ms, p99 29.835 ms, 479.8 QPS (1000/QPS 2.084 ms).
Passes in the order run: default@1, exact, default@8; each after its own warm-up. Untimed searches (rehearsals, per-pass warm-ups, settle checks and extensions): 65,908 sent, 0 failed (warmupErrors). Index after the searches: ready, 524 of 524 indexed (1 data part(s), index files on 1 of them, 524 of 524 live rows in indexed parts; 0 merge(s) running, 0 unfinished mutation(s); plan of the default search uses the Skip index vec_idx on 1 of 1 parts).
Benchmark copy gvbbench_eshoponweb dropped afterwards.
Connection: container address on its Docker network, not the published port (no docker-proxy): container gvb-clickhouse (ca31c0e3ce67) on network gvb-clickhouse_default at container-ip:8123, not through docker-proxy localhost:8123. Open after the passes: container-ip:8123 x8 (this target).
CPU pinning: cpuset from the container's creation when run-all started it on the engine CPUs, else docker update --cpuset-cpus (each container's change line says which), CPUs 2-3,6-7. Changed: container gvb-clickhouse: already on CPUs 2-3,6-7 (its cpuset since creation), not changed. Read back: container gvb-clickhouse: 655 thread(s) on CPUs 2-3,6-7 (every thread of the container).
Clock per pass (median MHz of engine CPUs / client CPUs, governor, CPUs busy outside the benchmark, client CPU per search): default@1 3590/3590 MHz, performance, outside load 0.14 before the warm-up, 0.16 during, client CPU 0.819 ms per search; exact 3592/3592 MHz, performance, outside load 0.14 before the warm-up, 0.16 during, client CPU 0.857 ms per search; default@8 3524/3539 MHz, performance, outside load 0.14 before the warm-up, 0.14 during, client CPU 1.023 ms per search.

Qdrant (exact) qdrant

Engine: Qdrant 1.17.0 in container gvb-qdrant (image qdrant/qdrant:v1.17.0@sha256:f1c7272cdac52b38c1a0e89313922d940ba50afd90d593a1605dbbc214e66ffb; cpuset 2-3,6-7 from its creation; 3 search thread(s)) (compose)
Index: exact scan: the builder's sink sends exact=true on every search, so no HNSW graph is used whether or not Qdrant has built one (see the index state)
Search settings: exact=true
Load: 524 rows in batches of 1000, 0.1 s of upserts (9,365 rows/s); count matched 0.0 s after the last upsert; index step 1.0 s (status Green, 0 of 524 vectors in HNSW segments (2 segments), waited 1.0 s)
Search: 22455 latency samples, target held 524 rows, 0 errors
Client CPU per search: 1 searcher 0.74 ms, 8 searchers 0.53 ms
RAM: 35.72 MiB (docker stats); disk: 196.08 MiB (collection folder)
Started by run-all for this measurement with every container created on CPUs 2-3,6-7 (started qdrant.compose.yaml with every container created on CPUs 2-3,6-7: gvb-qdrant cpuset 2-3,6-7 (read back from docker inspect)); stopped afterwards.
Index after the load: ready, 0 of 524 indexed (status Green, optimizer ok, indexed_vectors_count 0 of 524 points, 2 segments, indexing_threshold_kb 10,000, full_scan_threshold_kb 10,000; no index used, exact scan by design (the builder's sink sends exact=true on every search); no graph was built; searches counted since the collection was created: unfiltered_hnsw 0, unfiltered_plain 0, unfiltered_exact 0). Durability: What was measured (strace -f on the Qdrant server process, 30 single-point REST upserts per setting, every sync call timed against the request that caused it): with wait=true each upsert had exactly one msync(MS_SYNC) of the write-ahead-log segment inside the request, before the reply (30 of 30); with wait=false the 30 upserts were acknowledged within 0.26 s with no sync call, and the first WAL msync (one call covering all 30 records) came 2.7 s after the last reply, followed by the segment-file flushes. So the log is flushed to disk under both settings, but with wait=false the flush comes after the acknowledgement: an operating-system crash or power cut in that gap loses acknowledged writes, while a crash of the Qdrant process alone should not, because the bytes are already in the kernel's page cache (inferred, not tested). Every upsert here is sent with wait=true in batches of 256 over gRPC; the strace used one point per request, so one flush per batch is inferred, not measured. Segment files are flushed every 5 s; log segments 32 MB, 0 created ahead. /qdrant/config/config.yaml in container gvb-qdrant (the image's own file; its storage keys are listed) sets: storage.collection.quantization = null, storage.collection.replication_factor = 1, storage.collection.vectors.on_disk = null, storage.collection.write_consistency_factor = 1, storage.hnsw_index.ef_construct = 100, storage.hnsw_index.full_scan_threshold_kb = 10000, storage.hnsw_index.m = 16, storage.hnsw_index.max_indexing_threads = 0, storage.hnsw_index.on_disk = false, storage.hnsw_index.payload_m = null, storage.max_collections = null, storage.node_type = Normal, storage.on_disk_payload = true, storage.optimizers.default_segment_number = 0, storage.optimizers.deleted_threshold = 0.2, storage.optimizers.flush_interval_sec = 5, storage.optimizers.indexing_threshold_kb = 10000, storage.optimizers.max_optimization_threads = null, storage.optimizers.max_segment_size_kb = null, storage.optimizers.vacuum_min_vector_number = 1000, storage.performance.max_search_threads = 0, storage.performance.optimizer_cpu_budget = 0, storage.performance.update_rate_limit = null, storage.shard_transfer_method = null, storage.snapshots_config.snapshots_storage = local, storage.snapshots_path = ./snapshots, storage.storage_path = ./storage, storage.temp_path = null, storage.update_concurrency = null, storage.wal.wal_capacity_mb = 32, storage.wal.wal_segments_ahead = 0; no QDRANT__ environment overrides in the server process. Not tested by cutting power; whether the disk's own write cache reaches the media was not checked..
Load average 7.53 5.07 4.08 (1/5/15 min) when searching began.
Rehearsal before any timed pass, untimed, every pass type at its own concurrency for at least 30 s through the same code the timed passes use: default@8 98,326 searches in 30.0 s, default@1 33,514 searches in 30.0 s.
Settle: settled before every timed pass, each by its own warm-up at its own concurrency and a trial of the same pass within 10% of the warm-up's settled figure. default@8: warm-up 15.0 s and 49,361 searches (0 failed) at 8 searchers; QPS of the last windows 3,284, 3,270, 3,300 (windows of at least 2 s and 100 searches); trial of 9,901 searches in 3.0 s at 8 searchers: 3,298 QPS against the settled 3,284 QPS, 0% apart (limit 10%); default@1: warm-up 15.0 s and 16,739 searches (0 failed) at 1 searcher; p50 of the last windows 0.886, 0.880, 0.886 ms (windows of at least 2 s and 100 searches); trial of 3,350 searches in 3.0 s at 1 searcher: p50 0.882 ms against the settled p50 0.886 ms, 0% apart (limit 10%).
Pass default@8 after rehearsal: 2026-10-05T10:58:19.874Z to 2026-10-05T10:58:39.875Z (20.0 s), 65,915 searches, 0 failed, p50 2.338 ms, mean 2.425 ms, p99 4.226 ms, 3295.5 QPS (1000/QPS 0.303 ms).
Pass default@1 after default@8: 2026-10-05T10:58:57.902Z to 2026-10-05T10:59:17.903Z (20.0 s), 22,455 searches, 0 failed, p50 0.879 ms, mean 0.889 ms, p99 1.401 ms, 1122.7 QPS (1000/QPS 0.891 ms).
Passes in the order run: default@8, default@1; each after its own warm-up. Untimed searches (rehearsals, per-pass warm-ups, settle checks and extensions): 211,191 sent, 0 failed (warmupErrors). Index after the searches: ready, 0 of 524 indexed (status Green, optimizer ok, indexed_vectors_count 0 of 524 points, 2 segments, indexing_threshold_kb 10,000, full_scan_threshold_kb 10,000; no index used, exact scan by design (the builder's sink sends exact=true on every search); no graph was built; searches counted since the collection was created: unfiltered_hnsw 0, unfiltered_plain 599,122, unfiltered_exact 0).
Benchmark copy gvbbench_eshoponweb dropped afterwards.
Connection: container address on its Docker network, not the published port (no docker-proxy): container gvb-qdrant (dac2e8c05ea8) on network gvb-qdrant_default at container-ip:6334, not through docker-proxy localhost:16334; container gvb-qdrant (dac2e8c05ea8) on network gvb-qdrant_default at container-ip:6333, not through docker-proxy localhost:16333. Open after the passes: container-ip:6334 x1 (this target).
CPU pinning: cpuset from the container's creation when run-all started it on the engine CPUs, else docker update --cpuset-cpus (each container's change line says which), CPUs 2-3,6-7. Changed: container gvb-qdrant: already on CPUs 2-3,6-7 (its cpuset since creation), not changed. Read back: container gvb-qdrant: 24 thread(s) on CPUs 2-3,6-7 (every thread of the container).
Clock per pass (median MHz of engine CPUs / client CPUs, governor, CPUs busy outside the benchmark, client CPU per search): default@8 3492/3492 MHz, performance, outside load 0.13 before the warm-up, 0.14 during, client CPU 0.534 ms per search; default@1 3492/3492 MHz, performance, outside load 0.14 before the warm-up, 0.13 during, client CPU 0.736 ms per search.

pgvector

Engine: PostgreSQL 17 + pgvector 0.8.7 (compose)
Index: HNSW vector_cosine_ops m=16 ef_construction=128, hnsw.ef_search=100 per query, float32 vector(n), cosine; exact mode = same query with index scans off (sequential scan)
Search settings: ef_construction=128, hnsw.ef_search=100, m=16
Load: 524 rows in batches of 1000, 0.9 s of upserts (612 rows/s); count matched 0.0 s after the last upsert; index step 0.0 s (HNSW is maintained inside every insert, nothing to build; confirmed in 0.0 s: pg_indexes lists gvb_gvbbench_eshoponweb_hnsw (USING hnsw (embedding vector_cosine_ops) WITH (m='16', ef_construction='128')), pg_index valid and ready = True, EXPLAIN of the default search uses Index Scan on it = True, that search returned 10 of 10 rows; PostgreSQL keeps no entry count for an HNSW index, so indexed vectors are not reported)
Search: 23018 latency samples, target held 524 rows, 0 errors
Client CPU per search: 1 searcher 0.68 ms, 8 searchers 0.54 ms
Exact mode: 27,127 searches, p50 2.17 ms, p95 2.40 ms, recall 1.000
RAM: 100.1 MiB (docker stats); disk: 7.52 MiB added by this load (422.62 MiB (whole engine data folder), 415.1 MiB before)
Started by run-all for this measurement with every container created on CPUs 2-3,6-7 (started pgvector.compose.yaml with every container created on CPUs 2-3,6-7: gvb-pgvector cpuset 2-3,6-7 (read back from docker inspect)); stopped afterwards.
Index after the load: ready, ? of 524 indexed (pg_indexes lists gvb_gvbbench_eshoponweb_hnsw (USING hnsw (embedding vector_cosine_ops) WITH (m='16', ef_construction='128')), pg_index valid and ready = True, EXPLAIN of the default search uses Index Scan on it = True, that search returned 10 of 10 rows; PostgreSQL keeps no entry count for an HNSW index, so indexed vectors are not reported). Durability: fsync on, synchronous_commit on, full_page_writes on, wal_sync_method fdatasync (PostgreSQL defaults; pgvector.compose.yaml sets only shared_buffers, maintenance_work_mem and max_wal_size): every commit is flushed to the write-ahead log before it returns and HNSW index changes are WAL-logged, so a crash loses no committed row (max_wal_size 4GB only spaces out checkpoints; measured with docker kill, which is SIGKILL, right after a 2,000-vector load and a restart: all 2,000 rows were there and the HNSW index was valid and used by the default search).
Load average 3.80 4.81 4.14 (1/5/15 min) when searching began.
Rehearsal before any timed pass, untimed, every pass type at its own concurrency for at least 30 s through the same code the timed passes use: default@8 124,881 searches in 30.0 s, exact 13,453 searches in 30.0 s, default@1 34,836 searches in 30.0 s.
Settle: settled before every timed pass, each by its own warm-up at its own concurrency and a trial of the same pass within 10% of the warm-up's settled figure. default@8: warm-up 15.0 s and 60,638 searches (0 failed) at 8 searchers; QPS of the last windows 4,028, 4,086, 4,004 (windows of at least 2 s and 100 searches); trial of 12,098 searches in 3.0 s at 8 searchers: 4,031 QPS against the settled 4,028 QPS, 0% apart (limit 10%); exact: warm-up 15.0 s and 6,770 searches (0 failed) at 1 searcher; p50 of the last windows 2.185, 2.193, 2.159 ms (windows of at least 2 s and 100 searches); trial of 1,350 searches in 3.0 s at 1 searcher: p50 2.177 ms against the settled p50 2.185 ms, 0% apart (limit 10%); default@1: warm-up 15.0 s and 17,280 searches (0 failed) at 1 searcher; p50 of the last windows 0.829, 0.845, 0.856 ms (windows of at least 2 s and 100 searches); trial of 3,437 searches in 3.0 s at 1 searcher: p50 0.854 ms against the settled p50 0.845 ms, 1% apart (limit 10%).
Pass default@8 after rehearsal: 2026-10-05T11:01:15.252Z to 2026-10-05T11:01:35.253Z (20.0 s), 81,004 searches, 0 failed, p50 1.932 ms, mean 1.973 ms, p99 3.477 ms, 4050.0 QPS (1000/QPS 0.247 ms).
Pass exact after default@8: 2026-10-05T11:01:53.283Z to 2026-10-05T11:02:53.284Z (60.0 s), 27,127 searches, 0 failed, p50 2.174 ms, mean 2.210 ms, p99 2.976 ms, 452.1 QPS (1000/QPS 2.212 ms).
Pass default@1 after exact: 2026-10-05T11:03:11.313Z to 2026-10-05T11:03:31.313Z (20.0 s), 23,018 searches, 0 failed, p50 0.851 ms, mean 0.867 ms, p99 1.283 ms, 1150.9 QPS (1000/QPS 0.869 ms).
Passes in the order run: default@8, exact, default@1; each after its own warm-up. Untimed searches (rehearsals, per-pass warm-ups, settle checks and extensions): 274,743 sent, 0 failed (warmupErrors). Index after the searches: ready, ? of 524 indexed (pg_indexes lists gvb_gvbbench_eshoponweb_hnsw (USING hnsw (embedding vector_cosine_ops) WITH (m='16', ef_construction='128')), pg_index valid and ready = True, EXPLAIN of the default search uses Index Scan on it = True, that search returned 10 of 10 rows; PostgreSQL keeps no entry count for an HNSW index, so indexed vectors are not reported).
Benchmark copy gvbbench_eshoponweb dropped afterwards.
Connection: container address on its Docker network, not the published port (no docker-proxy): container gvb-pgvector (fcff93eb2a57) on network gvb-pgvector_default at container-ip:5432, not through docker-proxy localhost:5432. Open after the passes: container-ip:5432 x8 (this target).
CPU pinning: cpuset from the container's creation when run-all started it on the engine CPUs, else docker update --cpuset-cpus (each container's change line says which), CPUs 2-3,6-7. Changed: container gvb-pgvector: already on CPUs 2-3,6-7 (its cpuset since creation), not changed. Read back: container gvb-pgvector: 6 thread(s) on CPUs 2-3,6-7 (every thread of the container).
Clock per pass (median MHz of engine CPUs / client CPUs, governor, CPUs busy outside the benchmark, client CPU per search): default@8 3492/3492 MHz, performance, outside load 0.14 before the warm-up, 0.13 during, client CPU 0.539 ms per search; exact 3492/3492 MHz, performance, outside load 0.14 before the warm-up, 0.14 during, client CPU 0.672 ms per search; default@1 3492/3492 MHz, performance, outside load 0.14 before the warm-up, 0.13 during, client CPU 0.683 ms per search.

Vespa vespa

Engine: Vespa 8.754.14 (tensor attribute + HNSW) (compose)
Index: HNSW float32 tensor, prenormalized-angular (cosine), max-links-per-node=16, neighbors-to-explore-at-insert=128; search targetHits=top, ef=100 via exploreAdditionalHits; exact mode = approximate:false; vectors held in memory
Search settings: ef=100, max-links-per-node=16, neighbors-to-explore-at-insert=128, targetHits=top
Load: 524 rows in batches of 1000, 1.6 s of upserts (322 rows/s); count matched 0.0 s after the last upsert; index step 0.1 s (nothing to build, Vespa inserts into the HNSW graph while writing; ready after 0.1 s: HNSW graph returns 524 of 524 vectors: totalCount 524, coverage full True, nearestNeighbor approximate:true with targetHits above the count: has_index True, algorithm "index top k", top_k_hits 524; the graph is updated while each write is applied, so there is no later build step)
Search: 10540 latency samples, target held 524 rows, 0 errors
Client CPU per search: 1 searcher 0.92 ms, 8 searchers 0.86 ms
Exact mode: 31,578 searches, p50 1.84 ms, p95 2.28 ms, recall 1.000
RAM: 2.79 GiB (docker stats); disk: 12.11 MiB added by this load (2.12 GiB (whole engine data folder), 2.11 GiB before)
Started by run-all for this measurement with every container created on CPUs 2-3,6-7 (started vespa.compose.yaml with every container created on CPUs 2-3,6-7: gvb-vespa cpuset 2-3,6-7 (read back from docker inspect)); stopped afterwards.
Index after the load: ready, 524 of 524 indexed (HNSW graph returns 524 of 524 vectors: totalCount 524, coverage full True, nearestNeighbor approximate:true with targetHits above the count: has_index True, algorithm "index top k", top_k_hits 524; the graph is updated while each write is applied, so there is no later build step). Durability: Vespa's transaction log server fsyncs after each commit: searchlib.translogserver usefsync=true, and an operation is searchable when it is acknowledged: proton documentdb visibilitydelay=0. Neither is overridden by this sink or by the services.xml it deploys; VespaReadinessTests reads both from the live config server. After a crash the node replays the transaction log over its last flushed data and rebuilds the in-memory HNSW graph. By those settings an acknowledged write survives a crash; this is read from the settings, not shown by pulling power. One node and min-redundancy 1, so a lost disk loses the data..
Load average 6.58 4.69 4.24 (1/5/15 min) when searching began.
Rehearsal before any timed pass, untimed, every pass type at its own concurrency for at least 30 s through the same code the timed passes use: exact 9,270 searches in 30.0 s, default@8 37,287 searches in 30.0 s, default@1 15,679 searches in 30.0 s.
Settle: settled before every timed pass, each by its own warm-up at its own concurrency and a trial of the same pass within 10% of the warm-up's settled figure. exact: warm-up 15.0 s and 7,357 searches (0 failed) at 1 searcher; p50 of the last windows 1.893, 1.883, 1.859 ms (windows of at least 2 s and 100 searches); trial of 1,564 searches in 3.0 s at 1 searcher: p50 1.847 ms against the settled p50 1.883 ms, 2% apart (limit 10%); default@8: warm-up 15.0 s and 23,935 searches (0 failed) at 8 searchers; QPS of the last windows 1,612, 1,607, 1,548 (windows of at least 2 s and 100 searches); trial of 4,762 searches in 3.0 s at 8 searchers: 1,585 QPS against the settled 1,607 QPS, 1% apart (limit 10%); default@1: warm-up 15.0 s and 7,884 searches (0 failed) at 1 searcher; p50 of the last windows 1.843, 1.832, 1.844 ms (windows of at least 2 s and 100 searches); trial of 1,606 searches in 3.0 s at 1 searcher: p50 1.826 ms against the settled p50 1.843 ms, 1% apart (limit 10%).
Pass exact after rehearsal: 2026-10-05T11:06:14.663Z to 2026-10-05T11:07:14.663Z (60.0 s), 31,578 searches, 0 failed, p50 1.838 ms, mean 1.897 ms, p99 2.629 ms, 526.3 QPS (1000/QPS 1.900 ms).
Pass default@8 after exact: 2026-10-05T11:07:32.699Z to 2026-10-05T11:07:52.703Z (20.0 s), 31,910 searches, 0 failed, p50 4.882 ms, mean 5.012 ms, p99 8.332 ms, 1595.2 QPS (1000/QPS 0.627 ms).
Pass default@1 after default@8: 2026-10-05T11:08:10.715Z to 2026-10-05T11:08:30.716Z (20.0 s), 10,540 searches, 0 failed, p50 1.834 ms, mean 1.895 ms, p99 2.663 ms, 527.0 QPS (1000/QPS 1.898 ms).
Passes in the order run: exact, default@8, default@1; each after its own warm-up. Untimed searches (rehearsals, per-pass warm-ups, settle checks and extensions): 109,344 sent, 0 failed (warmupErrors). Index after the searches: ready, 524 of 524 indexed (HNSW graph returns 524 of 524 vectors: totalCount 524, coverage full True, nearestNeighbor approximate:true with targetHits above the count: has_index True, algorithm "index top k", top_k_hits 524; the graph is updated while each write is applied, so there is no later build step).
Benchmark copy gvbbench_eshoponweb dropped afterwards.
Connection: container address on its Docker network, not the published port (no docker-proxy): container gvb-vespa (204e1994e8c6) on network engines_default at container-ip:8080, not through docker-proxy localhost:8090; container gvb-vespa (204e1994e8c6) on network engines_default at container-ip:19071, not through docker-proxy localhost:19071. Open after the passes: container-ip:8080 x1 (this target).
CPU pinning: cpuset from the container's creation when run-all started it on the engine CPUs, else docker update --cpuset-cpus (each container's change line says which), CPUs 2-3,6-7. Changed: container gvb-vespa: already on CPUs 2-3,6-7 (its cpuset since creation), not changed. Read back: container gvb-vespa: 214 thread(s) on CPUs 2-3,6-7 (every thread of the container).
Clock per pass (median MHz of engine CPUs / client CPUs, governor, CPUs busy outside the benchmark, client CPU per search): exact 3520/3497 MHz, performance, outside load 0.16 before the warm-up, 0.12 during, client CPU 0.915 ms per search; default@8 3492/3493 MHz, performance, outside load 0.12 before the warm-up, 0.20 during, client CPU 0.863 ms per search; default@1 3513/3495 MHz, performance, outside load 0.17 before the warm-up, 0.10 during, client CPU 0.920 ms per search.

sqlite-vec sqlitevec

Engine: SQLite 3.53.3 + sqlite-vec 0.1.7-alpha.2.1 (vec0, embedded, in process) (embedded)
Index: vec0 brute-force scan, no ANN index (exact), float32, cosine distance, default chunk_size=1024; score = 1 - cosine distance; searches run concurrently, one WAL reader connection per searcher (opened as searchers arrive, at most 32), writes run one at a time and may overlap searches
Search settings: chunk_size=1024
Load: 524 rows in batches of 1000, 0.1 s of upserts (4,041 rows/s); count matched 0.0 s after the last upsert; index step 0.0 s (nothing to build, every search scans all vectors; checked in 0.00 s: no index, exact scan by design: gvb_gvbbench_eshoponweb is a vec0 virtual table holding 524 vectors and every search compares all of them (EXPLAIN QUERY PLAN: SCAN gvb_gvbbench_eshoponweb VIRTUAL TABLE INDEX 0:3{___}___ | USE TEMP B-TREE FOR ORDER BY))
Search: 10874 latency samples, target held 524 rows, 0 errors
Client CPU per search: 1 searcher 1.93 ms, 8 searchers 3.45 ms
Exact mode: 32,598 searches, p50 1.81 ms, p95 2.03 ms, recall 1.000
RAM: in this process, not measured; disk: 9.72 MiB added by this load (9.72 MiB (engine data folder), 0 B before)
Index after the load: ready, ? of 524 indexed (no index, exact scan by design: gvb_gvbbench_eshoponweb is a vec0 virtual table holding 524 vectors and every search compares all of them (EXPLAIN QUERY PLAN: SCAN gvb_gvbbench_eshoponweb VIRTUAL TABLE INDEX 0:3{___}___ | USE TEMP B-TREE FOR ORDER BY)). Durability: PRAGMA journal_mode=WAL and PRAGMA synchronous=NORMAL (set by this sink when it opens the file): each commit is appended to the -wal file and handed to the operating system, but the WAL is fsynced only when SQLite checkpoints it (measured with strace: 40 single-record commits caused 3 WAL fsyncs), so a crash of the process loses nothing (measured with kill -9 and a reopen: all 50, 300 and 2,000 rows were there), while an operating-system crash or power cut can lose the newest commits since the last checkpoint and leaves the database consistent.
Load average 4.83 5.09 4.53 (1/5/15 min) when searching began.
Rehearsal before any timed pass, untimed, every pass type at its own concurrency for at least 30 s through the same code the timed passes use: exact 16,153 searches in 30.0 s, default@8 34,747 searches in 30.0 s, default@1 16,329 searches in 30.0 s.
Settle: settled before every timed pass, each by its own warm-up at its own concurrency and a trial of the same pass within 10% of the warm-up's settled figure. exact: warm-up 15.0 s and 8,119 searches (0 failed) at 1 searcher; p50 of the last windows 1.822, 1.821, 1.802 ms (windows of at least 2 s and 100 searches); trial of 1,622 searches in 3.0 s at 1 searcher: p50 1.819 ms against the settled p50 1.821 ms, 0% apart (limit 10%); default@8: warm-up 15.0 s and 17,411 searches (0 failed) at 8 searchers; QPS of the last windows 1,159, 1,166, 1,159 (windows of at least 2 s and 100 searches); trial of 3,486 searches in 3.0 s at 8 searchers: 1,161 QPS against the settled 1,159 QPS, 0% apart (limit 10%); default@1: warm-up 15.0 s and 8,175 searches (0 failed) at 1 searcher; p50 of the last windows 1.804, 1.796, 1.821 ms (windows of at least 2 s and 100 searches); trial of 1,636 searches in 3.0 s at 1 searcher: p50 1.800 ms against the settled p50 1.804 ms, 0% apart (limit 10%).
Pass exact after rehearsal: 2026-10-05T11:10:28.432Z to 2026-10-05T11:11:28.433Z (60.0 s), 32,598 searches, 0 failed, p50 1.813 ms, mean 1.838 ms, p99 2.289 ms, 543.3 QPS (1000/QPS 1.841 ms).
Pass default@8 after exact: 2026-10-05T11:11:46.480Z to 2026-10-05T11:12:06.484Z (20.0 s), 23,183 searches, 0 failed, p50 6.702 ms, mean 6.893 ms, p99 13.605 ms, 1158.9 QPS (1000/QPS 0.863 ms).
Pass default@1 after default@8: 2026-10-05T11:12:24.494Z to 2026-10-05T11:12:44.495Z (20.0 s), 10,874 searches, 0 failed, p50 1.812 ms, mean 1.837 ms, p99 2.264 ms, 543.7 QPS (1000/QPS 1.839 ms).
Passes in the order run: exact, default@8, default@1; each after its own warm-up. Untimed searches (rehearsals, per-pass warm-ups, settle checks and extensions): 107,678 sent, 0 failed (warmupErrors). Index after the searches: ready, ? of 524 indexed (no index, exact scan by design: gvb_gvbbench_eshoponweb is a vec0 virtual table holding 524 vectors and every search compares all of them (EXPLAIN QUERY PLAN: SCAN gvb_gvbbench_eshoponweb VIRTUAL TABLE INDEX 0:3{___}___ | USE TEMP B-TREE FOR ORDER BY)).
Benchmark copy gvbbench_eshoponweb dropped afterwards.
Connection: in this process (embedded), no network: no network address. No open connection seen (no open TCP connection was found).
CPU pinning: none: an embedded engine runs inside the benchmark client process, so it shares the client CPUs, CPUs 0-1,4-5.
Clock per pass (median MHz of engine CPUs / client CPUs, governor, CPUs busy outside the benchmark, client CPU per search): exact 3597/3592 MHz, performance, outside load 0.16 before the warm-up, 0.15 during, client CPU 1.931 ms per search; default@8 3592/3592 MHz, performance, outside load 0.15 before the warm-up, 0.16 during, client CPU 3.446 ms per search; default@1 3592/3592 MHz, performance, outside load 0.16 before the warm-up, 0.16 during, client CPU 1.928 ms per search.

SQL Server 2025 + DiskANN sql-diskann

Engine: Microsoft SQL Server 2025 (RTM-CU9) (KB5122048) - 17.0.5005.3 (X64), Enterprise Developer Edition (64-bit), in container gvb-mssql (image mcr.microsoft.com/mssql/server:2025-CU9-ubuntu-24.04@sha256:2b5b581621126574f3d1f75e78d3eebe8d05aedb59ad0cfdf9aa42cb0634d726; cpuset 2-3,6-7 from its creation; SQL Server counts 4 CPU(s) and runs 4 visible scheduler(s), affinity AUTO) (compose)
Index: DiskANN (preview) via VECTOR_SEARCH, cosine, build {"StartId":"306", "L":"48", "M":"8", "R":"48"}; the exact mode scans the same table
Search settings: L=48, M=8, R=48, StartId=306
Load: 524 rows in batches of 1000, 0.8 s of upserts (630 rows/s); count matched 0.0 s after the last upsert; index step 3.9 s (copied 524 rows to dbo.gvb_gvbbench_eshoponweb_ann (INT key) and built DiskANN, parameters {"StartId":"306", "L":"48", "M":"8", "R":"48"}, graph covers 524 rows)
Search: 5468 latency samples, target held 524 rows, 0 errors
Client CPU per search: 1 searcher 1.01 ms, 8 searchers 1.07 ms
Exact mode: 16,386 searches, p50 3.58 ms, p95 4.14 ms, recall 1.000
RAM: 533 MiB (docker stats); disk: 4.52 MiB (table and its indexes, reserved pages)
Started by run-all for this measurement with every container created on CPUs 2-3,6-7 (started mssql.compose.yaml with every container created on CPUs 2-3,6-7: gvb-mssql cpuset 2-3,6-7 (read back from docker inspect)); stopped afterwards.
Index after the load: ready, 524 of 524 indexed (DiskANN index built and used. sys.vector_indexes: vix_gvb_gvbbench_eshoponweb on dbo.gvb_gvbbench_eshoponweb_ann in GvbBenchDiskAnn, DiskANN, COSINE, enabled, build {"StartId":"306", "L":"48", "M":"8", "R":"48"}; graph table rows 524 of 524 table rows. Real VECTOR_SEARCH plan: Vector Index Seek on index vix_gvb_gvbbench_eshoponweb of GvbBenchDiskAnn.gvb_gvbbench_eshoponweb_ann (IndexKind DiskANN), 10 rows returned, 10 hits. The exact mode searches the SAME table (dbo.gvb_gvbbench_eshoponweb_ann, database GvbBenchDiskAnn) and its real plan is: Clustered Index Scan of GvbBenchDiskAnn.gvb_gvbbench_eshoponweb_ann through PK_gvb_gvbbench_eshoponweb_ann (IndexKind Clustered), 524 rows read, no vector index operator, 10 hits). Durability: A commit returns after its transaction-log records are written to disk (SQL Server write-ahead logging; delayed durability is DISABLED); a database created here copies the model database: recovery model FULL, page_verify CHECKSUM; no global trace flags are enabled; mssql.conf of container gvb-mssql (/var/opt/mssql/mssql.conf, on the host at ~/gvb-data/engines/mssql/mssql.conf) does not exist, so it sets: nothing; it has no [control] or [traceflag] entry, so SQL Server's own Linux defaults for flushing writes apply (Microsoft's Linux performance guide names trace flag 3982 as that default; read from the guide, not tested here). Not tested by cutting power; whether the disk's own write cache reaches the media was not checked. Container settings from its environment (names only): MSSQL_AGENT_ENABLED, MSSQL_MEMORY_LIMIT_MB, MSSQL_PID, MSSQL_RPC_PORT, MSSQL_SA_PASSWORD..
Load average 2.68 3.89 4.17 (1/5/15 min) when searching began.
Rehearsal before any timed pass, untimed, every pass type at its own concurrency for at least 30 s through the same code the timed passes use: default@1 8,128 searches in 30.0 s, default@8 20,337 searches in 30.0 s, exact 8,153 searches in 30.0 s.
Settle: settled before every timed pass, each by its own warm-up at its own concurrency and a trial of the same pass within 10% of the warm-up's settled figure. default@1: warm-up 15.0 s and 4,050 searches (0 failed) at 1 searcher; p50 of the last windows 3.615, 3.629, 3.631 ms (windows of at least 2 s and 100 searches); trial of 823 searches in 3.0 s at 1 searcher: p50 3.620 ms against the settled p50 3.629 ms, 0% apart (limit 10%); default@8: warm-up 15.0 s and 10,322 searches (0 failed) at 8 searchers; QPS of the last windows 700, 698, 674 (windows of at least 2 s and 100 searches); trial of 2,066 searches in 3.1 s at 8 searchers: 675 QPS against the settled 698 QPS, 3% apart (limit 10%); exact: warm-up 15.0 s and 4,123 searches (0 failed) at 1 searcher; p50 of the last windows 3.548, 3.584, 3.562 ms (windows of at least 2 s and 100 searches); trial of 819 searches in 3.0 s at 1 searcher: p50 3.558 ms against the settled p50 3.562 ms, 0% apart (limit 10%).
Pass default@1 after rehearsal: 2026-10-05T11:14:45.772Z to 2026-10-05T11:15:05.775Z (20.0 s), 5,468 searches, 0 failed, p50 3.616 ms, mean 3.656 ms, p99 5.093 ms, 273.4 QPS (1000/QPS 3.658 ms).
Pass default@8 after default@1: 2026-10-05T11:15:23.851Z to 2026-10-05T11:15:43.861Z (20.0 s), 13,814 searches, 0 failed, p50 11.370 ms, mean 11.583 ms, p99 19.944 ms, 690.4 QPS (1000/QPS 1.449 ms).
Pass exact after default@8: 2026-10-05T11:16:01.869Z to 2026-10-05T11:17:01.870Z (60.0 s), 16,386 searches, 0 failed, p50 3.577 ms, mean 3.658 ms, p99 5.391 ms, 273.1 QPS (1000/QPS 3.662 ms).
Passes in the order run: default@1, default@8, exact; each after its own warm-up. Untimed searches (rehearsals, per-pass warm-ups, settle checks and extensions): 58,821 sent, 0 failed (warmupErrors). Index after the searches: ready, 524 of 524 indexed (DiskANN index built and used. sys.vector_indexes: vix_gvb_gvbbench_eshoponweb on dbo.gvb_gvbbench_eshoponweb_ann in GvbBenchDiskAnn, DiskANN, COSINE, enabled, build {"StartId":"306", "L":"48", "M":"8", "R":"48"}; graph table rows 524 of 524 table rows. Real VECTOR_SEARCH plan: Vector Index Seek on index vix_gvb_gvbbench_eshoponweb of GvbBenchDiskAnn.gvb_gvbbench_eshoponweb_ann (IndexKind DiskANN), 10 rows returned, 10 hits. The exact mode searches the SAME table (dbo.gvb_gvbbench_eshoponweb_ann, database GvbBenchDiskAnn) and its real plan is: Clustered Index Scan of GvbBenchDiskAnn.gvb_gvbbench_eshoponweb_ann through PK_gvb_gvbbench_eshoponweb_ann (IndexKind Clustered), 524 rows read, no vector index operator, 10 hits).
Benchmark copy gvbbench_eshoponweb dropped afterwards.
Connection: container address on its Docker network, not the published port (no docker-proxy): container gvb-mssql (7726b981139d) on network gvb-mssql_default at container-ip:1433, not through docker-proxy localhost:14330. Open after the passes: container-ip:1433 x9 (this target).
CPU pinning: cpuset from the container's creation when run-all started it on the engine CPUs, else docker update --cpuset-cpus (each container's change line says which), CPUs 2-3,6-7. Changed: container gvb-mssql: already on CPUs 2-3,6-7 (its cpuset since creation), not changed. Read back: container gvb-mssql: 85 thread(s) on CPUs 2-3,6-7 (every thread of the container).
Clock per pass (median MHz of engine CPUs / client CPUs, governor, CPUs busy outside the benchmark, client CPU per search): default@1 3497/3493 MHz, performance, outside load 0.15 before the warm-up, 0.15 during, client CPU 1.007 ms per search; default@8 3492/3492 MHz, performance, outside load 0.15 before the warm-up, 0.13 during, client CPU 1.068 ms per search; exact 3506/3494 MHz, performance, outside load 0.13 before the warm-up, 0.15 during, client CPU 1.118 ms per search.

Milvus milvus

Engine: Milvus 2.6.25 (standalone, embedded etcd, local storage, REST API v2) (compose)
Index: HNSW M=16 efConstruction=128, ef=100, metric COSINE, Strong consistency searches; approximate only (no exact mode)
Search settings: M=16, ef=100, efConstruction=128
Load: 524 rows in batches of 1000, 0.4 s of upserts (1,198 rows/s); count matched 1.6 s after the last upsert; index step 138.8 s (index Finished, type HNSW (COSINE, {"M":16,"efConstruction":128}), indexedRows 524 of 524 sealed rows, pendingRows 0; stored rows 524; LoadStateLoaded; query node: 1 sealed segment(s) with the index loaded covering 524 rows against 524 stored rows, 0 sealed without it, 0 rows in growing segments; datacoord: 1 live segment(s), 1 flushed and indexed covering 524 rows, 0 delete-log (L0) segment(s) waiting for a compaction, 3 segment(s) compacted away so far; flush, index build and load took 14.7 s, then compaction: 1 job(s) accepted by Milvus, layout ready and unchanged for 75 s, 124.1 s for the whole step; search ledger started, later state reads count the searches against the query node's own counters)
Search: 8576 latency samples, target held 524 rows, 0 errors
Client CPU per search: 1 searcher 0.54 ms, 8 searchers 0.63 ms
RAM: 215.5 MiB (docker stats); disk: 10.85 MiB added by this load (193.39 MiB (whole engine data folder), 182.53 MiB before)
Started by run-all for this measurement with every container created on CPUs 2-3,6-7 (started milvus.compose.yaml with every container created on CPUs 2-3,6-7: gvb-milvus cpuset 2-3,6-7 (read back from docker inspect)); stopped afterwards.
Index after the load: ready, 524 of 524 indexed (index Finished, type HNSW (COSINE, {"M":16,"efConstruction":128}), indexedRows 524 of 524 sealed rows, pendingRows 0; stored rows 524; LoadStateLoaded; query node: 1 sealed segment(s) with the index loaded covering 524 rows against 524 stored rows, 0 sealed without it, 0 rows in growing segments; datacoord: 1 live segment(s), 1 flushed and indexed covering 524 rows, 0 delete-log (L0) segment(s) waiting for a compaction, 3 segment(s) compacted away so far; search ledger since the index finished: the sink sent 0 searches (searchParams.params.ef=100, Strong consistency), the query node counted 0 search requests for this collection, segment searches Sealed +0 and Growing +0, the same sealed segments with the same loaded index builds at both readings, segments compacted away +0). Durability: Writes are not fsynced. Standalone uses the default message queue rocksmq (RocksDB, /var/lib/milvus/rdb_data, mq.type default) and flushed segments go to local-disk object storage (COMMON_STORAGETYPE=local in milvus.compose.yaml). Measured 2026-10-04 with strace over 30 acknowledged single-row upserts and one flush: fdatasync ran only on the embedded etcd files (member/wal, member/snap/db), never on rdb_data or the segment files. A host power loss or kernel crash can lose acknowledged rows that are still in the OS page cache; a Milvus process crash alone should not (inferred, not tested). Metadata in embedded etcd is fdatasynced on every commit..
Load average 0.36 2.05 3.30 (1/5/15 min) when searching began.
Rehearsal before any timed pass, untimed, every pass type at its own concurrency for at least 30 s through the same code the timed passes use: default@1 12,878 searches in 30.0 s, default@8 38,329 searches in 30.0 s.
Settle: settled before every timed pass, each by its own warm-up at its own concurrency and a trial of the same pass within 10% of the warm-up's settled figure. default@1: warm-up 15.0 s and 6,460 searches (0 failed) at 1 searcher; p50 of the last windows 2.219, 2.201, 2.204 ms (windows of at least 2 s and 100 searches); trial of 1,295 searches in 3.0 s at 1 searcher: p50 2.204 ms against the settled p50 2.204 ms, 0% apart (limit 10%); default@8: warm-up 15.0 s and 19,232 searches (0 failed) at 8 searchers; QPS of the last windows 1,288, 1,292, 1,280 (windows of at least 2 s and 100 searches); trial of 3,872 searches in 3.0 s at 8 searchers: 1,290 QPS against the settled 1,288 QPS, 0% apart (limit 10%).
Pass default@1 after rehearsal: 2026-10-05T11:20:52.933Z to 2026-10-05T11:21:12.935Z (20.0 s), 8,576 searches, 0 failed, p50 2.208 ms, mean 2.330 ms, p99 4.241 ms, 428.8 QPS (1000/QPS 2.332 ms).
Pass default@8 after default@1: 2026-10-05T11:21:30.948Z to 2026-10-05T11:21:50.952Z (20.0 s), 25,707 searches, 0 failed, p50 5.698 ms, mean 6.222 ms, p99 14.192 ms, 1285.1 QPS (1000/QPS 0.778 ms).
Passes in the order run: default@1, default@8; each after its own warm-up. Untimed searches (rehearsals, per-pass warm-ups, settle checks and extensions): 82,066 sent, 0 failed (warmupErrors). Index after the searches: ready, 524 of 524 indexed (index Finished, type HNSW (COSINE, {"M":16,"efConstruction":128}), indexedRows 524 of 524 sealed rows, pendingRows 0; stored rows 524; LoadStateLoaded; query node: 1 sealed segment(s) with the index loaded covering 524 rows against 524 stored rows, 0 sealed without it, 0 rows in growing segments; datacoord: 1 live segment(s), 1 flushed and indexed covering 524 rows, 0 delete-log (L0) segment(s) waiting for a compaction, 3 segment(s) compacted away so far; search ledger since the index finished: the sink sent 116349 searches (searchParams.params.ef=100, Strong consistency), the query node counted 116349 search requests for this collection, segment searches Sealed +114333 and Growing +0, the same sealed segments with the same loaded index builds at both readings, segments compacted away +0).
Benchmark copy gvbbench_eshoponweb dropped afterwards.
Connection: container address on its Docker network, not the published port (no docker-proxy): container gvb-milvus (e4f879f92338) on network gvb-milvus_default at container-ip:19530, not through docker-proxy localhost:19530; container gvb-milvus (e4f879f92338) on network gvb-milvus_default at container-ip:9091, not through docker-proxy localhost:9091. Open after the passes: container-ip:19530 x8 (this target).
CPU pinning: cpuset from the container's creation when run-all started it on the engine CPUs, else docker update --cpuset-cpus (each container's change line says which), CPUs 2-3,6-7. Changed: container gvb-milvus: already on CPUs 2-3,6-7 (its cpuset since creation), not changed. Read back: container gvb-milvus: 24 thread(s) on CPUs 2-3,6-7 (every thread of the container).
Clock per pass (median MHz of engine CPUs / client CPUs, governor, CPUs busy outside the benchmark, client CPU per search): default@1 3538/3506 MHz, performance, outside load 0.16 before the warm-up, 0.17 during, client CPU 0.544 ms per search; default@8 3492/3493 MHz, performance, outside load 0.17 before the warm-up, 0.18 during, client CPU 0.631 ms per search.

Elasticsearch elasticsearch

Engine: Elasticsearch 9.5.3 (dense_vector) (compose)
Index: HNSW float32, no quantization, m=16, ef_construction=128, cosine; search k=top, num_candidates=100; 1 shard, 0 replicas; force-merged to one segment after the load (at 1,024 dimensions a segment under 1,043 vectors gets no graph)
Search settings: ef_construction=128, k=top, m=16, num_candidates=100
Load: 524 rows in batches of 1000, 1.0 s of upserts (545 rows/s); count matched 0.2 s after the last upsert; index step 0.4 s (FAILED after 0.4 s: NO HNSW GRAPH, searches scan all 524 vectors: 1 segment(s), 524 vectors, total_vex_size_bytes 0 (_stats dense_vector), index_options hnsw m=16 ef_construction=128; Lucene builds no graph for a segment this small (measured at 1,024 dimensions: 1,042 vectors none, 1,043 vectors a graph), so default search equals exact search)
Search: 15012 latency samples, target held 524 rows, 0 errors
Client CPU per search: 1 searcher 0.45 ms, 8 searchers 0.49 ms
Exact mode: 46,958 searches, p50 1.26 ms, p95 1.46 ms, recall 1.000
RAM: 2.58 GiB (docker stats); disk: 2.39 MiB added by this load (2.55 MiB (whole engine data folder), 164.67 KiB before)
Started by run-all for this measurement with every container created on CPUs 2-3,6-7 (started elasticsearch.compose.yaml with every container created on CPUs 2-3,6-7: gvb-elasticsearch cpuset 2-3,6-7 (read back from docker inspect)); stopped afterwards.
WARNING: the index step after the load FAILED (NO HNSW GRAPH, searches scan all 524 vectors: 1 segment(s), 524 vectors, total_vex_size_bytes 0 (_stats dense_vector), index_options hnsw m=16 ef_construction=128; Lucene builds no graph for a segment this small (measured at 1,024 dimensions: 1,042 vectors none, 1,043 vectors a graph), so default search equals exact search). Searched anyway; see the index state.
WARNING: index not ready after the load: NOT ready, 0 of 524 indexed (NO HNSW GRAPH, searches scan all 524 vectors: 1 segment(s), 524 vectors, total_vex_size_bytes 0 (_stats dense_vector), index_options hnsw m=16 ef_construction=128; Lucene builds no graph for a segment this small (measured at 1,024 dimensions: 1,042 vectors none, 1,043 vectors a graph), so default search equals exact search). Measured anyway, so its search numbers may come from a scan or a half-built index. Durability: Every acknowledged bulk request is fsynced to the translog before the answer: index.translog.durability=request, the Elasticsearch default, which neither this sink nor elasticsearch.compose.yaml overrides (ElasticsearchReadinessTests reads it back from the live index). By that setting a process crash or power loss loses no acknowledged write; this is read from the setting, not shown by pulling power. One node and no replicas, so a lost disk loses the data..
Load average 3.38 3.02 3.47 (1/5/15 min) when searching began.
Rehearsal before any timed pass, untimed, every pass type at its own concurrency for at least 30 s through the same code the timed passes use: exact 15,233 searches in 30.0 s, default@8 61,862 searches in 30.0 s, default@1 22,492 searches in 30.0 s.
Settle: settled before every timed pass, each by its own warm-up at its own concurrency and a trial of the same pass within 10% of the warm-up's settled figure. exact: warm-up 15.0 s and 11,494 searches (0 failed) at 1 searcher; p50 of the last windows 1.260, 1.250, 1.258 ms (windows of at least 2 s and 100 searches); trial of 2,343 searches in 3.0 s at 1 searcher: p50 1.258 ms against the settled p50 1.258 ms, 0% apart (limit 10%); default@8: warm-up 15.0 s and 40,299 searches (0 failed) at 8 searchers; QPS of the last windows 2,668, 2,672, 2,726 (windows of at least 2 s and 100 searches); trial of 8,115 searches in 3.0 s at 8 searchers: 2,703 QPS against the settled 2,672 QPS, 1% apart (limit 10%); default@1: warm-up 15.0 s and 11,254 searches (0 failed) at 1 searcher; p50 of the last windows 1.317, 1.306, 1.317 ms (windows of at least 2 s and 100 searches); trial of 2,239 searches in 3.0 s at 1 searcher: p50 1.313 ms against the settled p50 1.317 ms, 0% apart (limit 10%).
Pass exact after rehearsal: 2026-10-05T11:24:15.986Z to 2026-10-05T11:25:15.986Z (60.0 s), 46,958 searches, 0 failed, p50 1.255 ms, mean 1.276 ms, p99 1.707 ms, 782.6 QPS (1000/QPS 1.278 ms).
Pass default@8 after exact: 2026-10-05T11:25:34.037Z to 2026-10-05T11:25:54.038Z (20.0 s), 53,954 searches, 0 failed, p50 2.834 ms, mean 2.963 ms, p99 5.434 ms, 2697.5 QPS (1000/QPS 0.371 ms).
Pass default@1 after default@8: 2026-10-05T11:26:12.059Z to 2026-10-05T11:26:32.060Z (20.0 s), 15,012 searches, 0 failed, p50 1.312 ms, mean 1.330 ms, p99 1.759 ms, 750.6 QPS (1000/QPS 1.332 ms).
Passes in the order run: exact, default@8, default@1; each after its own warm-up. Untimed searches (rehearsals, per-pass warm-ups, settle checks and extensions): 175,331 sent, 0 failed (warmupErrors). Index after the searches: NOT ready, 0 of 524 indexed (NO HNSW GRAPH, searches scan all 524 vectors: 1 segment(s), 524 vectors, total_vex_size_bytes 0 (_stats dense_vector), index_options hnsw m=16 ef_construction=128; Lucene builds no graph for a segment this small (measured at 1,024 dimensions: 1,042 vectors none, 1,043 vectors a graph), so default search equals exact search).
Benchmark copy gvbbench_eshoponweb dropped afterwards.
Connection: container address on its Docker network, not the published port (no docker-proxy): container gvb-elasticsearch (082b04c91b12) on network engines_default at container-ip:9200, not through docker-proxy localhost:9200. Open after the passes: container-ip:9200 x8 (this target).
CPU pinning: cpuset from the container's creation when run-all started it on the engine CPUs, else docker update --cpuset-cpus (each container's change line says which), CPUs 2-3,6-7. Changed: container gvb-elasticsearch: already on CPUs 2-3,6-7 (its cpuset since creation), not changed. Read back: container gvb-elasticsearch: 74 thread(s) on CPUs 2-3,6-7 (every thread of the container).
Clock per pass (median MHz of engine CPUs / client CPUs, governor, CPUs busy outside the benchmark, client CPU per search): exact 3492/3492 MHz, performance, outside load 0.16 before the warm-up, 0.15 during, client CPU 0.453 ms per search; default@8 3492/3492 MHz, performance, outside load 0.15 before the warm-up, 0.21 during, client CPU 0.489 ms per search; default@1 3492/3492 MHz, performance, outside load 0.18 before the warm-up, 0.14 during, client CPU 0.449 ms per search.

Redis redis

Engine: Redis 8.10.2 (query engine, HASH + vector index) (compose)
Index: HNSW TYPE FLOAT32 M=16 EF_CONSTRUCTION=128, EF_RUNTIME=100 per query, cosine; exact mode = FLAT index built on first exact query
Search settings: EF_CONSTRUCTION=128, EF_RUNTIME=100, M=16
Load: 524 rows in batches of 1000, 0.0 s of upserts (11,939 rows/s); count matched 0.0 s after the last upsert; index step 0.0 s (HNSW index finished after 0.0 s of waiting: FT.INFO indexing 0, percent_indexed 1, hash_indexing_failures 0, flat_buffer_size 0, num_docs 524, hashes counted with SCAN under gvb_gvbbench_eshoponweb: 524)
Search: 51896 latency samples, target held 524 rows, 0 errors
Client CPU per search: 1 searcher 0.55 ms, 8 searchers 0.37 ms
Exact mode: 165,985 searches, p50 0.36 ms, p95 0.45 ms, recall 1.000
RAM: 21.19 MiB (docker stats); disk: 2.43 MiB added by this load (2.43 MiB (whole engine data folder), 89 B before)
Started by run-all for this measurement with every container created on CPUs 2-3,6-7 (started redis.compose.yaml with every container created on CPUs 2-3,6-7: gvb-redis cpuset 2-3,6-7 (read back from docker inspect)); stopped afterwards.
Index after the load: ready, 524 of 524 indexed (FT.INFO indexing 0, percent_indexed 1, hash_indexing_failures 0, flat_buffer_size 0, num_docs 524, hashes counted with SCAN under gvb_gvbbench_eshoponweb: 524). Durability: save "300 1" and appendonly no (redis.compose.yaml): an RDB snapshot is written every 5 minutes if at least one key changed, and on a clean stop; there is no append-only log, so a crash loses every write since the last snapshot (up to 5 minutes plus the time a snapshot takes; measured with docker kill, which is SIGKILL, right after a 2,000-vector load: none of the 2,000 hashes were there after the restart, and the restart loaded an older snapshot that still held keys of a collection that had been dropped since).
Load average 2.63 3.12 3.43 (1/5/15 min) when searching began.
Rehearsal before any timed pass, untimed, every pass type at its own concurrency for at least 30 s through the same code the timed passes use: exact 83,254 searches in 30.0 s, default@1 79,550 searches in 30.0 s, default@8 150,945 searches in 30.0 s.
WARNING: latency had NOT settled when timing began for exact (each by its own warm-up at its own concurrency and a trial of the same pass within 10% of the warm-up's settled figure). exact: warm-up 15.0 s and 41,314 searches (0 failed) at 1 searcher; p50 of the last windows 0.336, 0.374, 0.372 ms (windows of at least 2 s and 100 searches) (its last 3 windows differed by more than 5%); trial of 8,416 searches in 3.0 s at 1 searcher: p50 0.335 ms against the settled p50 0.372 ms, 11% apart (limit 10%), so the warm-up was EXTENDED once (30.0 s and 81,883 searches (0 failed) at 1 searcher; p50 of the last windows 0.363, 0.369, 0.338 ms (windows of at least 2 s and 100 searches), its last 3 windows differed by more than 5%); second trial of 8,081 searches in 3.0 s at 1 searcher: p50 0.369 ms against the settled p50 0.363 ms, 2% apart (limit 10%); still disagreeing after the extension; default@1: warm-up 15.0 s and 38,516 searches (0 failed) at 1 searcher; p50 of the last windows 0.372, 0.407, 0.378 ms (windows of at least 2 s and 100 searches) (its last 3 windows differed by more than 5%); trial of 7,842 searches in 3.0 s at 1 searcher: p50 0.377 ms against the settled p50 0.378 ms, 0% apart (limit 10%), so the warm-up was EXTENDED once (32.0 s and 82,511 searches (0 failed) at 1 searcher; p50 of the last windows 0.385, 0.380, 0.391 ms (windows of at least 2 s and 100 searches)); second trial of 7,681 searches in 3.0 s at 1 searcher: p50 0.390 ms against the settled p50 0.385 ms, 1% apart (limit 10%); settled after the extension; default@8: warm-up 15.0 s and 75,402 searches (0 failed) at 8 searchers; QPS of the last windows 4,844, 5,008, 5,010 (windows of at least 2 s and 100 searches); trial of 14,794 searches in 3.0 s at 8 searchers: 4,929 QPS against the settled 5,008 QPS, 2% apart (limit 10%). Its numbers may still include warm-up; rerun before quoting them.
Pass exact after rehearsal: 2026-10-05T11:29:18.811Z to 2026-10-05T11:30:18.811Z (60.0 s), 165,985 searches, 0 failed, p50 0.358 ms, mean 0.360 ms, p99 0.704 ms, 2766.4 QPS (1000/QPS 0.361 ms).
Pass default@1 after exact: 2026-10-05T11:31:27.059Z to 2026-10-05T11:31:47.059Z (20.0 s), 51,896 searches, 0 failed, p50 0.374 ms, mean 0.384 ms, p99 0.707 ms, 2594.8 QPS (1000/QPS 0.385 ms).
Pass default@8 after default@1: 2026-10-05T11:32:05.122Z to 2026-10-05T11:32:25.123Z (20.0 s), 100,789 searches, 0 failed, p50 1.570 ms, mean 1.586 ms, p99 2.249 ms, 5039.1 QPS (1000/QPS 0.198 ms).
Passes in the order run: exact, default@1, default@8; each after its own warm-up. Untimed searches (rehearsals, per-pass warm-ups, settle checks and extensions): 760,524 sent, 0 failed (warmupErrors). Index after the searches: ready, 524 of 524 indexed (FT.INFO indexing 0, percent_indexed 1, hash_indexing_failures 0, flat_buffer_size 0, num_docs 524, hashes counted with SCAN under gvb_gvbbench_eshoponweb: 524).
Benchmark copy gvbbench_eshoponweb dropped afterwards.
Connection: container address on its Docker network, not the published port (no docker-proxy): container gvb-redis (755e6fb9597a) on network gvb-redis_default at container-ip:6379, not through docker-proxy localhost:6379. Open after the passes: container-ip:6379 x1 (this target).
CPU pinning: cpuset from the container's creation when run-all started it on the engine CPUs, else docker update --cpuset-cpus (each container's change line says which), CPUs 2-3,6-7. Changed: container gvb-redis: already on CPUs 2-3,6-7 (its cpuset since creation), not changed. Read back: container gvb-redis: 6 thread(s) on CPUs 2-3,6-7 (every thread of the container).
Clock per pass (median MHz of engine CPUs / client CPUs, governor, CPUs busy outside the benchmark, client CPU per search): exact 3492/3492 MHz, performance, outside load 0.14 before the warm-up, 0.14 during, client CPU 0.557 ms per search; default@1 3492/3492 MHz, performance, outside load 0.13 before the warm-up, 0.13 during, client CPU 0.554 ms per search; default@8 3492/3492 MHz, performance, outside load 0.13 before the warm-up, 0.14 during, client CPU 0.368 ms per search.

Weaviate weaviate

Engine: Weaviate 1.39.8 (single node, REST + GraphQL) (compose)
Index: HNSW maxConnections(M)=16 efConstruction=128, ef=-1 (dynamic: limit x 8 clamped 100..500), cosine, no quantization; approximate only (no exact mode)
Search settings: ef=-1, efConstruction=128
Load: 524 rows in batches of 1000, 0.5 s of upserts (1,018 rows/s); count matched 0.0 s after the last upsert; index step 0.0 s (HNSW is updated inside every batch, nothing to build; confirmed in 0.0 s: shard KRx7woimLViV vectorIndexingStatus READY, vectorQueueLength 0, node status objectCount 0 (refreshed only when the memtable is flushed, so it lags a minute and does not decide readiness), schema shard KRx7woimLViV status READY, Aggregate count 524, Aggregate nearVector with objectLimit 524 reached 524, Weaviate reports no count of vectors in the HNSW index, so indexed vectors are not reported)
Search: 3426 latency samples, target held 524 rows, 0 errors
Client CPU per search: 1 searcher 0.59 ms, 8 searchers 0.60 ms
RAM: 69.82 MiB (docker stats); disk: 2.96 MiB added by this load (5.02 MiB (whole engine data folder), 2.06 MiB before)
Started by run-all for this measurement with every container created on CPUs 2-3,6-7 (started weaviate.compose.yaml with every container created on CPUs 2-3,6-7: gvb-weaviate cpuset 2-3,6-7 (read back from docker inspect)); stopped afterwards.
Index after the load: ready, ? of 524 indexed (shard KRx7woimLViV vectorIndexingStatus READY, vectorQueueLength 0, node status objectCount 0 (refreshed only when the memtable is flushed, so it lags a minute and does not decide readiness), schema shard KRx7woimLViV status READY, Aggregate count 524, Aggregate nearVector with objectLimit 524 reached 524, Weaviate reports no count of vectors in the HNSW index, so indexed vectors are not reported). Durability: Weaviate 1.39.8 defaults, weaviate.compose.yaml sets no persistence variable [Correction 1]: every object is appended to the LSM write-ahead log with a plain write and no fsync, and the log is fsynced only when its memtable is flushed, 60 seconds after the last write (PERSISTENCE_MEMTABLES_FLUSH_IDLE_AFTER_SECONDS default; measured with strace: 2,080 writes into the objects log during a 2,000-object load, the first fsync 60 s after the last write), so a power cut can lose the last minute of writes; the HNSW commit log is buffered inside the process (77 writes for 2,000 vectors), so a killed process loses the newest vectors from the vector index while their objects survive (measured with docker kill, which is SIGKILL, one second after the load and a restart, two runs each: 0 to 1 of 50, 264 to 267 of 300 and 1,992 to 1,997 of 2,000 stored objects were still reachable through a vector search).
Load average 2.40 2.63 3.09 (1/5/15 min) when searching began.
Rehearsal before any timed pass, untimed, every pass type at its own concurrency for at least 30 s through the same code the timed passes use: default@8 9,471 searches in 30.0 s, default@1 5,179 searches in 30.0 s.
Settle: settled before every timed pass, each by its own warm-up at its own concurrency and a trial of the same pass within 10% of the warm-up's settled figure. default@8: warm-up 15.0 s and 4,694 searches (0 failed) at 8 searchers; QPS of the last windows 312, 309, 312 (windows of at least 2 s and 100 searches); trial of 943 searches in 3.0 s at 8 searchers: 312 QPS against the settled 312 QPS, 0% apart (limit 10%); default@1: warm-up 15.0 s and 2,568 searches (0 failed) at 1 searcher; p50 of the last windows 5.184, 5.190, 5.193 ms (windows of at least 2 s and 100 searches); trial of 515 searches in 3.0 s at 1 searcher: p50 5.194 ms against the settled p50 5.190 ms, 0% apart (limit 10%).
Pass default@8 after rehearsal: 2026-10-05T11:34:07.163Z to 2026-10-05T11:34:27.186Z (20.0 s), 6,250 searches, 0 failed, p50 24.398 ms, mean 25.624 ms, p99 52.729 ms, 312.1 QPS (1000/QPS 3.204 ms).
Pass default@1 after default@8: 2026-10-05T11:34:45.198Z to 2026-10-05T11:35:05.205Z (20.0 s), 3,426 searches, 0 failed, p50 5.192 ms, mean 5.837 ms, p99 11.434 ms, 171.2 QPS (1000/QPS 5.840 ms).
Passes in the order run: default@8, default@1; each after its own warm-up. Untimed searches (rehearsals, per-pass warm-ups, settle checks and extensions): 23,370 sent, 0 failed (warmupErrors). Index after the searches: ready, ? of 524 indexed (shard KRx7woimLViV vectorIndexingStatus READY, vectorQueueLength 0, node status objectCount 524 (refreshed only when the memtable is flushed, so it lags a minute and does not decide readiness), schema shard KRx7woimLViV status READY, Aggregate count 524, Aggregate nearVector with objectLimit 524 reached 524, Weaviate reports no count of vectors in the HNSW index, so indexed vectors are not reported).
Benchmark copy gvbbench_eshoponweb dropped afterwards.
Connection: container address on its Docker network, not the published port (no docker-proxy): container gvb-weaviate (92ede008fc8a) on network gvb-weaviate_default at container-ip:8080, not through docker-proxy localhost:8085. Open after the passes: container-ip:8080 x1 (this target).
CPU pinning: cpuset from the container's creation when run-all started it on the engine CPUs, else docker update --cpuset-cpus (each container's change line says which), CPUs 2-3,6-7. Changed: container gvb-weaviate: already on CPUs 2-3,6-7 (its cpuset since creation), not changed. Read back: container gvb-weaviate: 9 thread(s) on CPUs 2-3,6-7 (every thread of the container).
Clock per pass (median MHz of engine CPUs / client CPUs, governor, CPUs busy outside the benchmark, client CPU per search): default@8 3570/3562 MHz, performance, outside load 0.15 before the warm-up, 0.15 during, client CPU 0.599 ms per search; default@1 3585/3552 MHz, performance, outside load 0.16 before the warm-up, 0.14 during, client CPU 0.587 ms per search.

Oracle 23ai Free oracle

Engine: Oracle AI Database Free 23.26 (23ai line, VECTOR FLOAT32) (compose)
Index: HNSW in-memory neighbor graph NEIGHBORS=16 EFCONSTRUCTION=128, EFSEARCH=100 per query, cosine; exact mode = FETCH EXACT FIRST (full scan); Oracle Free caps itself at 2 CPUs (cpu_count 2 in V$PARAMETER, edition FREE in V$INSTANCE, 8 host CPUs in V$OSSTAT NUM_CPUS; the 2 CPU thread limit is Oracle's documented Free edition limit)
Search settings: EFCONSTRUCTION=128, EFSEARCH=100, NEIGHBORS=16
Load: 524 rows in batches of 1000, 0.3 s of upserts (2,020 rows/s); count matched 0.1 s after the last upsert; index step 3.5 s (HNSW graph holds 524 of 524 rows, change log waiting: 0 inserts and 0 deletes, USER_INDEXES status VALID, plan of the default search: VECTOR INDEX HNSW SCAN > TABLE ACCESS BY INDEX ROWID, index used by 0 queries so far; Oracle Free caps itself at 2 CPUs (cpu_count 2 in V$PARAMETER, edition FREE in V$INSTANCE, 8 host CPUs in V$OSSTAT NUM_CPUS; the 2 CPU thread limit is Oracle's documented Free edition limit); graph rebuilt in 2.2 s)
Search: 21936 latency samples, target held 524 rows, 0 errors
Client CPU per search: 1 searcher 0.87 ms, 8 searchers 0.65 ms
Exact mode: 24,710 searches, p50 2.36 ms, p95 2.85 ms, recall 1.000
RAM: 2.14 GiB (docker stats); disk: 0 B added by this load (6.88 GiB (whole engine data folder), 6.88 GiB before)
Started by run-all for this measurement with every container created on CPUs 2-3,6-7 (started oracle.compose.yaml with every container created on CPUs 2-3,6-7: gvb-oracle cpuset 2-3,6-7, gvb-oracle-seed cpuset 2-3,6-7 (read back from docker inspect)); stopped afterwards.
Index after the load: ready, 524 of 524 indexed (HNSW graph holds 524 of 524 rows, change log waiting: 0 inserts and 0 deletes, USER_INDEXES status VALID, plan of the default search: VECTOR INDEX HNSW SCAN > TABLE ACCESS BY INDEX ROWID, index used by 0 queries so far; Oracle Free caps itself at 2 CPUs (cpu_count 2 in V$PARAMETER, edition FREE in V$INSTANCE, 8 host CPUs in V$OSSTAT NUM_CPUS; the 2 CPU thread limit is Oracle's documented Free edition limit)). Durability: By these settings a committed row should survive a crash or power loss. The sink uses a plain COMMIT and commit_logging, commit_wait and commit_write are unset in the database, so every commit waits for its redo to be written. Measured 2026-10-04: 50 separate client commits raised V$SYSSTAT 'redo synch writes' by 55 (the 50 commits plus the CREATE and DROP of the probe table), and the log writer and the datafile writer hold their files open with O_DSYNC (open flags 02110002, filesystemio_options none). The database runs NOARCHIVELOG (V$DATABASE.LOG_MODE), so redo serves crash recovery only and there is no point-in-time restore. The HNSW graph lives in the 768 MB vector memory pool (oracle-init/01-vector-memory.sh) and is not the durable copy; the table is. Not tested by cutting power; whether the disk's own write cache reaches the media was not checked. CPU: Oracle Free caps itself at 2 CPUs (measured 2026-10-04: cpu_count 2 in V$PARAMETER, edition FREE in V$INSTANCE, 8 host CPUs in V$OSSTAT NUM_CPUS; the 2 CPU thread limit is Oracle's documented Free edition limit; this sink reads the live value when the index state is read)..
Load average 2.56 2.81 3.10 (1/5/15 min) when searching began.
Rehearsal before any timed pass, untimed, every pass type at its own concurrency for at least 30 s through the same code the timed passes use: exact 12,597 searches in 30.0 s, default@1 36,374 searches in 30.0 s, default@8 97,139 searches in 30.0 s.
WARNING: latency had NOT settled when timing began for default@8 (each by its own warm-up at its own concurrency and a trial of the same pass within 10% of the warm-up's settled figure). exact: warm-up 15.0 s and 6,064 searches (0 failed) at 1 searcher; p50 of the last windows 2.371, 2.350, 2.331 ms (windows of at least 2 s and 100 searches); trial of 1,244 searches in 3.0 s at 1 searcher: p50 2.356 ms against the settled p50 2.350 ms, 0% apart (limit 10%); default@1: warm-up 15.0 s and 16,495 searches (0 failed) at 1 searcher; p50 of the last windows 0.888, 0.870, 0.904 ms (windows of at least 2 s and 100 searches); trial of 3,251 searches in 3.0 s at 1 searcher: p50 0.896 ms against the settled p50 0.888 ms, 1% apart (limit 10%); default@8: warm-up 15.3 s and 44,258 searches (0 failed) at 8 searchers; QPS of the last windows 3,062, 2,834, 2,622 (windows of at least 2 s and 100 searches) (its last 3 windows differed by more than 5%); trial of 9,357 searches in 3.0 s at 8 searchers: 3,118 QPS against the settled 2,834 QPS, 9% apart (limit 10%), so the warm-up was EXTENDED once (120.2 s and 344,269 searches (0 failed) at 8 searchers; QPS of the last windows 3,394, 2,474, 2,998 (windows of at least 2 s and 100 searches), the 120 s cap ran out); second trial of 9,022 searches in 3.0 s at 8 searchers: 3,006 QPS against the settled 2,998 QPS, 0% apart (limit 10%); still disagreeing after the extension. Its numbers may still include warm-up; rerun before quoting them.
Pass exact after rehearsal: 2026-10-05T11:37:12.549Z to 2026-10-05T11:38:12.551Z (60.0 s), 24,710 searches, 0 failed, p50 2.360 ms, mean 2.426 ms, p99 3.922 ms, 411.8 QPS (1000/QPS 2.428 ms).
Pass default@1 after exact: 2026-10-05T11:38:30.577Z to 2026-10-05T11:38:50.578Z (20.0 s), 21,936 searches, 0 failed, p50 0.886 ms, mean 0.910 ms, p99 1.856 ms, 1096.8 QPS (1000/QPS 0.912 ms).
Pass default@8 after default@1: 2026-10-05T11:41:27.175Z to 2026-10-05T11:41:47.315Z (20.1 s), 56,923 searches, 0 failed, p50 1.606 ms, mean 2.828 ms, p99 3.498 ms, 2826.3 QPS (1000/QPS 0.354 ms).
Passes in the order run: exact, default@1, default@8; each after its own warm-up. Untimed searches (rehearsals, per-pass warm-ups, settle checks and extensions): 619,250 sent, 0 failed (warmupErrors). Index after the searches: ready, 524 of 524 indexed (HNSW graph holds 524 of 524 rows, change log waiting: 0 inserts and 0 deletes, USER_INDEXES status VALID, plan of the default search: VECTOR INDEX HNSW SCAN > TABLE ACCESS BY INDEX ROWID, index used by 678204 queries so far; Oracle Free caps itself at 2 CPUs (cpu_count 2 in V$PARAMETER, edition FREE in V$INSTANCE, 8 host CPUs in V$OSSTAT NUM_CPUS; the 2 CPU thread limit is Oracle's documented Free edition limit)).
Benchmark copy gvbbench_eshoponweb dropped afterwards.
Connection: container address on its Docker network, not the published port (no docker-proxy): container gvb-oracle (e8b2151b656f) on network gvb-oracle_default at container-ip:1521, not through docker-proxy localhost:1521. Open after the passes: container-ip:1521 x14 (this target).
CPU pinning: cpuset from the container's creation when run-all started it on the engine CPUs, else docker update --cpuset-cpus (each container's change line says which), CPUs 2-3,6-7. Changed: container gvb-oracle: already on CPUs 2-3,6-7 (its cpuset since creation), not changed. Read back: container gvb-oracle: 91 thread(s) on CPUs 2-3,6-7 (every thread of the container).
Clock per pass (median MHz of engine CPUs / client CPUs, governor, CPUs busy outside the benchmark, client CPU per search): exact 3492/3492 MHz, performance, outside load 0.17 before the warm-up, 0.16 during, client CPU 0.930 ms per search; default@1 3492/3492 MHz, performance, outside load 0.16 before the warm-up, 0.16 during, client CPU 0.870 ms per search; default@8 3492/3492 MHz, performance, outside load 0.16 before the warm-up, 0.16 during, client CPU 0.649 ms per search.

Notes

Order: targets ran one at a time in a random order from seed 603 (runSeed; --seed 603 repeats it, targetOrder lists it). Inside each target the timed passes also ran in a random order from the same seed and the target's name (passOrder): default@1 is one searcher for 20 s (every search's latency gives p50/p95/p99, the completed searches give QPS@1, its first answer to each query gives recall and nDCG), default@N is N searchers for 20 s (QPS@N), exact is the engine's exact mode, one searcher for 60 s cycling the queries.
Preparation, untimed, before any timed pass: a rehearsal of every pass type at its own concurrency for 30 s each, through the same code the passes use.
Warm-up and settle check, untimed, right before every timed pass: the pass's own search at the pass's own number of searchers for at least 15 s and at least 20 searches (at most 120 s), read in windows of at least 2 s and 100 searches; then a 3 s trial of the same pass. The trial's figure (p50 with one searcher, QPS with several) must lie within 10% of the warm-up's settled figure (the median of its last 3 windows, which must agree within 5%); if not, the warm-up is extended once (at least 30 s, until its windows agree, at most 120 s), a second trial is taken and the warm-up runs again before the pass. Each target's notes give every check, and a pass that still disagrees flags its target as unsettled. With machine control on, the check for a quiet box is made when the warm-up is announced, before it starts (a wait between the warm-up and the timed pass let the engine go cold), so by the time the clock opens that check is as old as the warm-up and the trial (about 18 s, more after an extension); each pass's conditions record that lead (quietCheckLeadSeconds). Failed untimed searches are counted per target (warmupErrors) and are not in the timed error counts.
Load rows/s counts only time inside each target's upsert calls: one writer, batches of 1000, rows already in memory, the collection dropped and created fresh first. Engines that build or finish their index after the writes do it in a separate timed index step (the load's index seconds), and the run waits for it before searching.
Index proof: each engine's own report of its index (indexState) is read after the load and again after the last pass. A target whose index was not ready after the load was still measured and carries a WARNING; an engine that reports nothing counts as not ready.
Search settings: each searched target records the settings its own index description states (searchSettings: the build parameters and the effort per query, such as m, ef_construction and ef_search), read after the load; consolidating never averages runs whose settings differ.
Durability: each target's crash-safety setting as configured here (durability). Engines that do not force writes to disk on every commit load faster for that reason.
Latency is client-side wall time around each search (network and driver included), every search of the default@1 window, one searcher, the queries cycled; a window with fewer than 200 searches, a p50 above 1.25 x its mean, or a mean above its p99 is flagged.
QPS: N searchers back to back for 20 s per level; completed searches divided by the window's elapsed time.
Recall@10: share of the exact top 10 (brute force in memory) that the engine returned. A hit whose exact similarity ties the 10th best (within 1e-5) also counts, because duplicate rows embed to identical vectors.
Routes: a container engine (the benchmark's own SQL Server and Qdrant containers included) is reached at its container's own address on its Docker network, never through the published localhost port (docker-proxy); the native comparison targets (sql-native, qdrant-native) are native services reached directly over loopback. Each target's addresses and the connections the client held open after its passes are in its notes and in conditions.connections.
RAM of compose engines, the benchmark's own SQL Server and Qdrant containers (sql, sql-diskann, qdrant, qdrant-hnsw) included, is docker stats of the engine's containers; for the native comparison targets (sql-native, qdrant-native) it is the whole native process, including every other database or collection it serves. Disk is the table's reserved pages (SQL), the collection folder (Qdrant), or for other engines what the load added to the engine's data folder (engines that keep data in memory until a snapshot show almost nothing).
Engines: run-all starts an engine that is down and stops it afterwards only if it was not running when the run began; an engine that was already running is left running.
WARNING: the engine did not report a ready index after the load for: elasticsearch. They were measured anyway; their numbers may come from a scan or a half-built index (see each target's index state).
Machine control on: governor performance on every CPU during the run (before: schedutil; at the end: performance; after putting it back: schedutil). engine CPUs 2-3,6-7 (cores 2,6 and 3,7), client CPUs 0-1,4-5 (cores 0,4 and 1,5); the client process was pinned; each engine was pinned to the engine CPUs for its turn and put back after (conditions.engines); an engine run-all started (and stopped) was asked to be created on them, and its notes say whether the host did so or it was moved there after the start. Busy box: right before each pass's warm-up the run waits, up to 10 min, while processes outside the benchmark (everything but this client and the engine under test's cgroups) use more than 0.3 CPUs on average over the last 60 s or the last 5 s, neither window reaching back past the start of the target or of its engine (so the engine's own start-up is not outside work); if the box does not clear the pass runs anyway, flagged 'busy box', as is a pass whose own outside load is above the limit. Outside load counts busy = user + nice + system + irq + softirq + steal (guest time is already in user); CONFIG_IRQ_TIME_ACCOUNTING is not set, so task and cgroup run time include the interrupt and softirq time that hit them and it is added; CONFIG_PARAVIRT_TIME_ACCOUNTING is not set, so steal is added (/boot/config-6.8.0-142-generic) (conditions.cpuAccounting); each pass also records this client's own CPU time per search (conditions.passes[].clientCpuMsPerSearch). CPU clocks were sampled every 250 ms; each pass's min/median/max per CPU is in conditions.passes. CPU idle states, recorded and left as found: driver intel_idle, governor menu, intel_idle max_cstate 9; POLL on, C1 on, C1E on, C3 OFF on CPUs 0-7 (default disabled), C6 on (conditions.cpuIdle). How each target was reached: conditions.connections. Client build Release, .NET 10.0.12.
duckdb is embedded: it ran inside the client process on the client CPUs 0-1,4-5, sharing them with the client.
sqlitevec is embedded: it ran inside the client process on the client CPUs 0-1,4-5, sharing them with the client.

Raw files

results.md is kept unedited. It still holds the sentences this page leaves out of its copy. Count: 2

The raw files below hold the recorded statements that the corrections above refer to, as written.