# Vector engine benchmark: eshoponweb

Run 2026-10-05T03:57:11Z (run-all). Command line:

```
~/gvb-work/lanes/v5-final/src/GenericVectorBuilder.Bench/bin/Release/net10.0/GenericVectorBuilder.Bench.dll run-all --pipeline eshoponweb --queries golden --seed 503 --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 3.86 3.17 2.88
- 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; 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. Where it is close to the latency, the client library is a large part of what is measured.
- 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 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| sqlitevec | vec0 brute-force scan, no ANN index (exact), float32, cosine distance, default chunk_size=1024; score = 1 - cosine distance | 4,031 | 1.83 | 2.02 | 2.26 | 539.9 | 532.7 | 1.95 | 1.98 | 1.000 | 0.518 | - | 9.72 MiB |
| sql | exact VECTOR_DISTANCE cosine, no vector index (full scan) | 570 | 3.90 | 4.41 | 5.22 | 251.2 | 622.8 | 1.21 | 1.17 | 1.000 | 0.518 | 527.9 MiB | 5.15 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) | 1,845 | 0.88 | 1.06 | 1.74 | 1101.8 | 3135.9 | 0.87 | 0.65 | 1.000 | 0.518 | 2.13 GiB | 0 B |
| 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 | 5,304 | 0.91 | 1.02 | 1.37 | 1082.4 | 3477.5 | 0.74 | 0.52 | 1.000 | 0.518 | 36.79 MiB | 164.1 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) | 967 | 5.20 | 10.4 | 11.4 | 172.7 | 315.1 | 0.59 | 0.60 | 1.000 | 0.518 | 75.93 MiB | 2.88 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 | 11,246 | 0.35 | 0.46 | 0.67 | 2735.6 | 5043.1 | 0.51 | 0.36 | 1.000 | 0.518 | 21.3 MiB | 0 B |
| 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)), mhnsw_max_cache_size 4G; exact mode = IGNORE INDEX full scan | 1,011 | 0.62 | 0.70 | 0.88 | 1579.9 | 5941.7 | 0.49 | 0.38 | 1.000 | 0.518 | 171 MiB | 22.01 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 | 644 | 3.56 | 4.02 | 5.29 | 276.9 | 694.2 | 1.00 | 1.07 | 0.965 | 0.526 | 530.1 MiB | 4.52 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 | 375 | 1.97 | 2.20 | 2.68 | 495.3 | 1446.2 | 0.47 | 0.51 | 1.000 | 0.518 | 2.6 GiB | 9.44 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) | 7,760 | 0.88 | 0.99 | 1.39 | 1125.8 | 3302.3 | 0.73 | 0.53 | 1.000 | 0.518 | 37.21 MiB | 196.08 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) | 560 | 1.31 | 1.46 | 1.74 | 750.3 | 2334.7 | 0.47 | 0.50 | 1.000 | 0.518 | 2.54 GiB | 2.38 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 | 1,603 | 3.50 | 3.92 | 4.97 | 282.8 | 279.2 | 4.82 | 4.94 | 1.000 | 0.518 | - | 2.9 MiB |
| chroma | HNSW M=16 ef_construction=128, ef_search=100 (Chroma default), cosine; approximate only (no exact mode) | 707 | 2.04 | 2.28 | 2.52 | 485.9 | 806.0 | 0.45 | 0.46 | 1.000 | 0.518 | 42.68 MiB | 414.16 KiB |
| 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 | 262 | 1.90 | 2.42 | 2.89 | 507.8 | 977.3 | 0.95 | 0.93 | 1.000 | 0.518 | 2.83 GiB | 6.39 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 | 494 | 4.05 | 4.36 | 4.65 | 244.6 | 588.0 | 0.51 | 0.57 | 1.000 | 0.518 | 118.3 MiB | 0 B |
| mongodb | vectorSearch index, HNSW maxEdges=16 numEdgeCandidates=128, float32 binData, cosine, numCandidates=20x hits (min 100); exact mode = $vectorSearch exact:true | 3,315 | 1.38 | 1.60 | 1.95 | 709.2 | 2038.1 | 0.31 | 0.28 | 1.000 | 0.518 | 680.4 MiB | 0 B |
| 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) | 623 | 0.83 | 0.98 | 1.24 | 1176.2 | 4120.5 | 0.69 | 0.54 | 1.000 | 0.518 | 100.4 MiB | 7.59 MiB |
| 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 | 2,882 | 5.20 | 7.78 | 9.06 | 176.6 | 413.1 | 0.90 | 1.04 | 0.990 | 0.528 | 738.5 MiB | 2.42 GiB |
| milvus | HNSW M=16 efConstruction=128, ef=100, metric COSINE, Strong consistency searches; approximate only (no exact mode) | 1,188 | 2.24 | 2.77 | 3.72 | 426.1 | 1260.5 | 0.55 | 0.64 | 1.000 | 0.518 | 198.4 MiB | 10.86 MiB |

## Details per target

### 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
- Load: 524 rows in batches of 1000, 0.1 s of upserts (4,031 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: 10798 latency samples, target held 524 rows, 0 errors
- Client CPU per search: 1 searcher 1.95 ms, 8 searchers 1.98 ms
- Exact mode: 32,355 searches, p50 1.83 ms, p95 2.04 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 3.86 3.17 2.88 (1/5/15 min) when searching began.
- Rehearsal before any timed pass, untimed, every pass type for at least 5 s through the same code the timed passes use: exact 2,570 searches in 5.0 s, default@8 2,665 searches in 5.0 s, default@1 2,672 searches in 5.0 s.
- Settle: settled after 0.6 s and 300 searches (0 failed); p50 of the last windows of 100: 1.81, 1.82, 1.81 ms, within 5% across 3 windows; confirmed by a trial right before the first timed pass of 1,058 searches in 2.0 s: p50 1.855 ms against the settled p50 1.812 ms, 2% apart (limit 15%).
- Pass exact after settle: 2026-10-05T03:57:31.369Z to 2026-10-05T03:58:31.370Z (60.0 s), 32,355 searches, 0 failed, p50 1.828 ms, mean 1.852 ms, p99 2.249 ms, 539.2 QPS (1000/QPS 1.854 ms).
- Pass default@8 after exact: 2026-10-05T03:58:31.464Z to 2026-10-05T03:58:51.479Z (20.0 s), 10,662 searches, 0 failed, p50 14.861 ms, mean 15.010 ms, p99 18.267 ms, 532.7 QPS (1000/QPS 1.877 ms).
- Pass default@1 after default@8: 2026-10-05T03:58:51.521Z to 2026-10-05T03:59:11.522Z (20.0 s), 10,798 searches, 0 failed, p50 1.829 ms, mean 1.850 ms, p99 2.259 ms, 539.9 QPS (1000/QPS 1.852 ms).
- Passes in the order run: exact, default@8, default@1; each after its own warm-up. Untimed searches (rehearsal, settle, settle check and per-pass warm-ups): 9,325 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. Open after the passes: localhost:1433 x2 (not this target).
- 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 0.18 before the warm-up, 0.13 during, client CPU 1.949 ms per search; default@8 3600/3592 MHz, performance, outside load 0.13 before the warm-up, 0.13 during, client CPU 1.982 ms per search; default@1 3600/3592 MHz, performance, outside load 0.12 before the warm-up, 0.12 during, client CPU 1.945 ms per search.

### 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)
- Load: 524 rows in batches of 1000, 0.9 s of upserts (570 rows/s); count matched 0.0 s after the last upsert; index step n/a
- Search: 5024 latency samples, target held 524 rows, 0 errors
- Client CPU per search: 1 searcher 1.21 ms, 8 searchers 1.17 ms
- RAM: 527.9 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 1.53 2.47 2.66 (1/5/15 min) when searching began.
- Rehearsal before any timed pass, untimed, every pass type for at least 5 s through the same code the timed passes use: default@8 2,715 searches in 5.0 s, default@1 1,246 searches in 5.0 s.
- Settle: settled after 1.2 s and 300 searches (0 failed); p50 of the last windows of 100: 3.87, 3.85, 3.86 ms, within 5% across 3 windows; confirmed by a trial right before the first timed pass of 479 searches in 2.0 s: p50 3.869 ms against the settled p50 3.863 ms, 0% apart (limit 15%).
- Pass default@8 after settle: 2026-10-05T03:59:35.590Z to 2026-10-05T03:59:55.599Z (20.0 s), 12,461 searches, 0 failed, p50 12.464 ms, mean 12.841 ms, p99 19.477 ms, 622.8 QPS (1000/QPS 1.606 ms).
- Pass default@1 after default@8: 2026-10-05T03:59:55.683Z to 2026-10-05T04:00:15.683Z (20.0 s), 5,024 searches, 0 failed, p50 3.902 ms, mean 3.978 ms, p99 5.216 ms, 251.2 QPS (1000/QPS 3.981 ms).
- Passes in the order run: default@8, default@1; each after its own warm-up. Untimed searches (rehearsal, settle, settle check and per-pass warm-ups): 4,780 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 (522bb917862c) 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@8 3492/3492 MHz, performance, outside load 0.15 before the warm-up, 0.14 during, client CPU 1.173 ms per search; default@1 3510/3494 MHz, performance, outside load 0.14 before the warm-up, 0.14 during, client CPU 1.205 ms per search.

### 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$OSSTAT NUM_CPUS; the 2 CPU thread limit is Oracle's documented Free edition limit)
- Load: 524 rows in batches of 1000, 0.3 s of upserts (1,845 rows/s); count matched 0.1 s after the last upsert; index step 3.4 s (HNSW graph holds 524 of 524 rows, change log waiting: 0 inserts and 0 deletes, USER_INDEXES status VALID, plan of the default search: VECTOR INDEX HNSW SCAN > TABLE ACCESS BY INDEX ROWID, index used by 0 queries so far; Oracle Free caps itself at 2 CPUs (cpu_count 2 in V$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: 22037 latency samples, target held 524 rows, 0 errors
- Client CPU per search: 1 searcher 0.87 ms, 8 searchers 0.65 ms
- Exact mode: 24,675 searches, p50 2.37 ms, p95 2.80 ms, recall 1.000
- RAM: 2.13 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 1.79 2.43 2.63 (1/5/15 min) when searching began.
- Rehearsal before any timed pass, untimed, every pass type for at least 5 s through the same code the timed passes use: default@8 13,778 searches in 5.1 s, exact 2,018 searches in 5.0 s, default@1 5,509 searches in 5.0 s.
- Settle: settled after 0.3 s and 300 searches (0 failed); p50 of the last windows of 100: 0.89, 0.92, 0.91 ms, within 5% across 3 windows; confirmed by a trial right before the first timed pass of 2,217 searches in 2.0 s: p50 0.876 ms against the settled p50 0.907 ms, 4% apart (limit 15%).
- Pass default@8 after settle: 2026-10-05T04:00:52.450Z to 2026-10-05T04:01:12.451Z (20.0 s), 62,722 searches, 0 failed, p50 1.588 ms, mean 2.549 ms, p99 3.403 ms, 3135.9 QPS (1000/QPS 0.319 ms).
- Pass exact after default@8: 2026-10-05T04:01:12.523Z to 2026-10-05T04:02:12.525Z (60.0 s), 24,675 searches, 0 failed, p50 2.373 ms, mean 2.430 ms, p99 3.734 ms, 411.2 QPS (1000/QPS 2.432 ms).
- Pass default@1 after exact: 2026-10-05T04:02:12.568Z to 2026-10-05T04:02:32.568Z (20.0 s), 22,037 searches, 0 failed, p50 0.884 ms, mean 0.906 ms, p99 1.737 ms, 1101.8 QPS (1000/QPS 0.908 ms).
- Passes in the order run: default@8, exact, default@1; each after its own warm-up. Untimed searches (rehearsal, settle, settle check and per-pass warm-ups): 23,882 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 106603 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 (bfb6a3d1fa25) 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: 86 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 0.12 before the warm-up, 0.12 during, client CPU 0.650 ms per search; exact 3492/3492 MHz, performance, outside load 0.12 before the warm-up, 0.11 during, client CPU 0.939 ms per search; default@1 3492/3492 MHz, performance, outside load 0.11 before the warm-up, 0.11 during, client CPU 0.869 ms per search.

### 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
- Load: 524 rows in batches of 1000, 0.1 s of upserts (5,304 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: 21649 latency samples, target held 524 rows, 0 errors
- Client CPU per search: 1 searcher 0.74 ms, 8 searchers 0.52 ms
- Exact mode: 66,683 searches, p50 0.89 ms, p95 1.00 ms, recall 1.000
- RAM: 36.79 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 1.88 2.48 2.64 (1/5/15 min) when searching began.
- Rehearsal before any timed pass, untimed, every pass type for at least 5 s through the same code the timed passes use: default@8 16,478 searches in 5.0 s, exact 5,550 searches in 5.0 s, default@1 5,407 searches in 5.0 s.
- Settle: settled after 0.3 s and 300 searches (0 failed); p50 of the last windows of 100: 0.91, 0.91, 0.91 ms, within 5% across 3 windows; confirmed by a trial right before the first timed pass of 2,173 searches in 2.0 s: p50 0.911 ms against the settled p50 0.914 ms, 0% apart (limit 15%).
- Pass default@8 after settle: 2026-10-05T04:03:03.230Z to 2026-10-05T04:03:23.232Z (20.0 s), 69,558 searches, 0 failed, p50 2.216 ms, mean 2.298 ms, p99 3.947 ms, 3477.5 QPS (1000/QPS 0.288 ms).
- Pass exact after default@8: 2026-10-05T04:03:23.276Z to 2026-10-05T04:04:23.276Z (60.0 s), 66,683 searches, 0 failed, p50 0.888 ms, mean 0.898 ms, p99 1.359 ms, 1111.4 QPS (1000/QPS 0.900 ms).
- Pass default@1 after exact: 2026-10-05T04:04:23.359Z to 2026-10-05T04:04:43.360Z (20.0 s), 21,649 searches, 0 failed, p50 0.914 ms, mean 0.922 ms, p99 1.374 ms, 1082.4 QPS (1000/QPS 0.924 ms).
- Passes in the order run: default@8, exact, default@1; each after its own warm-up. Untimed searches (rehearsal, settle, settle check and per-pass warm-ups): 29,968 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; 115,605 segment searches walked the graph and none scanned; searches counted since the collection was created: unfiltered_hnsw 115,605, unfiltered_plain 0, unfiltered_exact 72,253).
- Benchmark copy gvbbench_eshoponweb dropped afterwards.
- Connection: container address on its Docker network, not the published port (no docker-proxy): container gvb-qdrant (a7f938ab343f) on network gvb-qdrant_default at container-ip:6334, not through docker-proxy localhost:16334; container gvb-qdrant (a7f938ab343f) 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 0.09 before the warm-up, 0.11 during, client CPU 0.520 ms per search; exact 3492/3492 MHz, performance, outside load 0.10 before the warm-up, 0.08 during, client CPU 0.733 ms per search; default@1 3492/3492 MHz, performance, outside load 0.08 before the warm-up, 0.07 during, client CPU 0.736 ms per search.

### 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)
- Load: 524 rows in batches of 1000, 0.5 s of upserts (967 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 X7dNkH77J5pr 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 X7dNkH77J5pr 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: 3456 latency samples, target held 524 rows, 0 errors
- Client CPU per search: 1 searcher 0.59 ms, 8 searchers 0.60 ms
- RAM: 75.93 MiB (docker stats); disk: 2.88 MiB added by this load (4.94 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 X7dNkH77J5pr 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 X7dNkH77J5pr 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: 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 1.58 2.36 2.59 (1/5/15 min) when searching began.
- Rehearsal before any timed pass, untimed, every pass type for at least 5 s through the same code the timed passes use: default@8 1,584 searches in 5.0 s, default@1 861 searches in 5.0 s.
- Settle: settled after 1.7 s and 300 searches (0 failed); p50 of the last windows of 100: 5.21, 5.16, 5.19 ms, within 5% across 3 windows; confirmed by a trial right before the first timed pass of 346 searches in 2.0 s: p50 5.184 ms against the settled p50 5.186 ms, 0% apart (limit 15%).
- Pass default@8 after settle: 2026-10-05T04:05:21.150Z to 2026-10-05T04:05:41.171Z (20.0 s), 6,308 searches, 0 failed, p50 23.956 ms, mean 25.377 ms, p99 52.711 ms, 315.1 QPS (1000/QPS 3.174 ms).
- Pass default@1 after default@8: 2026-10-05T04:05:41.289Z to 2026-10-05T04:06:01.298Z (20.0 s), 3,456 searches, 0 failed, p50 5.197 ms, mean 5.787 ms, p99 11.437 ms, 172.7 QPS (1000/QPS 5.790 ms).
- Passes in the order run: default@8, default@1; each after its own warm-up. Untimed searches (rehearsal, settle, settle check and per-pass warm-ups): 3,131 sent, 0 failed (warmupErrors). Index after the searches: ready, ? of 524 indexed (shard X7dNkH77J5pr 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 X7dNkH77J5pr 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 (52c49c858939) 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 3553/3561 MHz, performance, outside load 0.12 before the warm-up, 0.11 during, client CPU 0.604 ms per search; default@1 3591/3556 MHz, performance, outside load 0.11 before the warm-up, 0.12 during, client CPU 0.587 ms per search.

### 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
- Load: 524 rows in batches of 1000, 0.0 s of upserts (11,246 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: 54713 latency samples, target held 524 rows, 0 errors
- Client CPU per search: 1 searcher 0.51 ms, 8 searchers 0.36 ms
- Exact mode: 177,804 searches, p50 0.32 ms, p95 0.43 ms, recall 1.000
- RAM: 21.3 MiB (docker stats); disk: 0 B added by this load (89 B (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 2.33 2.46 2.61 (1/5/15 min) when searching began.
- Rehearsal before any timed pass, untimed, every pass type for at least 5 s through the same code the timed passes use: exact 13,152 searches in 5.0 s, default@1 13,714 searches in 5.0 s, default@8 25,384 searches in 5.0 s.
- Settle: settled after 0.4 s and 1,100 searches (0 failed); p50 of the last windows of 100: 0.35, 0.34, 0.35 ms, within 5% across 3 windows; confirmed by a trial right before the first timed pass of 5,469 searches in 2.0 s: p50 0.346 ms against the settled p50 0.348 ms, 0% apart (limit 15%).
- Pass exact after settle: 2026-10-05T04:06:27.100Z to 2026-10-05T04:07:27.100Z (60.0 s), 177,804 searches, 0 failed, p50 0.318 ms, mean 0.336 ms, p99 0.635 ms, 2963.4 QPS (1000/QPS 0.337 ms).
- Pass default@1 after exact: 2026-10-05T04:07:27.284Z to 2026-10-05T04:07:47.285Z (20.0 s), 54,713 searches, 0 failed, p50 0.346 ms, mean 0.364 ms, p99 0.666 ms, 2735.6 QPS (1000/QPS 0.366 ms).
- Pass default@8 after default@1: 2026-10-05T04:07:47.347Z to 2026-10-05T04:08:07.348Z (20.0 s), 100,870 searches, 0 failed, p50 1.565 ms, mean 1.585 ms, p99 2.206 ms, 5043.1 QPS (1000/QPS 0.198 ms).
- Passes in the order run: exact, default@1, default@8; each after its own warm-up. Untimed searches (rehearsal, settle, settle check and per-pass warm-ups): 58,879 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 (9defcf1b7c3a) 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 0.09 before the warm-up, 0.09 during, client CPU 0.508 ms per search; default@1 3492/3492 MHz, performance, outside load 0.09 before the warm-up, 0.11 during, client CPU 0.512 ms per search; default@8 3492/3492 MHz, performance, outside load 0.10 before the warm-up, 0.10 during, client CPU 0.361 ms per search.

### 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)), mhnsw_max_cache_size 4G; exact mode = IGNORE INDEX full scan
- Load: 524 rows in batches of 1000, 0.5 s of upserts (1,011 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 394 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 394 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: 31599 latency samples, target held 524 rows, 0 errors
- Client CPU per search: 1 searcher 0.49 ms, 8 searchers 0.38 ms
- Exact mode: 37,284 searches, p50 1.59 ms, p95 1.68 ms, recall 1.000
- RAM: 171 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 (mariadb.compose.yaml): 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, because gvb-mariadb is shared and stays up).
- Load average 2.22 2.41 2.58 (1/5/15 min) when searching began.
- Rehearsal before any timed pass, untimed, every pass type for at least 5 s through the same code the timed passes use: default@1 7,819 searches in 5.0 s, exact 3,102 searches in 5.0 s, default@8 30,111 searches in 5.0 s.
- Settle: settled after 0.2 s and 400 searches (0 failed); p50 of the last windows of 100: 0.60, 0.59, 0.60 ms, within 5% across 3 windows; confirmed by a trial right before the first timed pass of 3,153 searches in 2.0 s: p50 0.626 ms against the settled p50 0.601 ms, 4% apart (limit 15%).
- Pass default@1 after settle: 2026-10-05T04:08:33.810Z to 2026-10-05T04:08:53.810Z (20.0 s), 31,599 searches, 0 failed, p50 0.622 ms, mean 0.631 ms, p99 0.883 ms, 1579.9 QPS (1000/QPS 0.633 ms).
- Pass exact after default@1: 2026-10-05T04:08:53.874Z to 2026-10-05T04:09:53.874Z (60.0 s), 37,284 searches, 0 failed, p50 1.585 ms, mean 1.608 ms, p99 2.030 ms, 621.4 QPS (1000/QPS 1.609 ms).
- Pass default@8 after exact: 2026-10-05T04:09:53.911Z to 2026-10-05T04:10:13.912Z (20.0 s), 118,840 searches, 0 failed, p50 1.303 ms, mean 1.345 ms, p99 2.470 ms, 5941.7 QPS (1000/QPS 0.168 ms).
- Passes in the order run: default@1, exact, default@8; each after its own warm-up. Untimed searches (rehearsal, settle, settle check and per-pass warm-ups): 44,645 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 (d77e3427a653) 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: 13 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 0.10 before the warm-up, 0.11 during, client CPU 0.487 ms per search; exact 3592/3592 MHz, performance, outside load 0.10 before the warm-up, 0.10 during, client CPU 0.494 ms per search; default@8 3492/3492 MHz, performance, outside load 0.10 before the warm-up, 0.07 during, client CPU 0.383 ms per search.

### 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
- Load: 524 rows in batches of 1000, 0.8 s of upserts (644 rows/s); count matched 0.0 s after the last upsert; index step 4.1 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: 5538 latency samples, target held 524 rows, 0 errors
- Client CPU per search: 1 searcher 1.00 ms, 8 searchers 1.07 ms
- Exact mode: 16,404 searches, p50 3.59 ms, p95 4.10 ms, recall 1.000
- RAM: 530.1 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 3.31 2.61 2.62 (1/5/15 min) when searching began.
- Rehearsal before any timed pass, untimed, every pass type for at least 5 s through the same code the timed passes use: default@8 3,100 searches in 5.0 s, exact 1,366 searches in 5.0 s, default@1 1,371 searches in 5.0 s.
- Settle: settled after 1.1 s and 300 searches (0 failed); p50 of the last windows of 100: 3.58, 3.61, 3.62 ms, within 5% across 3 windows; confirmed by a trial right before the first timed pass of 559 searches in 2.0 s: p50 3.564 ms against the settled p50 3.607 ms, 1% apart (limit 15%).
- Pass default@8 after settle: 2026-10-05T04:10:46.110Z to 2026-10-05T04:11:06.120Z (20.0 s), 13,891 searches, 0 failed, p50 11.313 ms, mean 11.519 ms, p99 19.966 ms, 694.2 QPS (1000/QPS 1.440 ms).
- Pass exact after default@8: 2026-10-05T04:11:06.197Z to 2026-10-05T04:12:06.199Z (60.0 s), 16,404 searches, 0 failed, p50 3.590 ms, mean 3.655 ms, p99 4.690 ms, 273.4 QPS (1000/QPS 3.658 ms).
- Pass default@1 after exact: 2026-10-05T04:12:06.283Z to 2026-10-05T04:12:26.286Z (20.0 s), 5,538 searches, 0 failed, p50 3.563 ms, mean 3.610 ms, p99 5.291 ms, 276.9 QPS (1000/QPS 3.612 ms).
- Passes in the order run: default@8, exact, default@1; each after its own warm-up. Untimed searches (rehearsal, settle, settle check and per-pass warm-ups): 6,756 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 (bd248cfd1c35) 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: 86 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 0.12 before the warm-up, 0.10 during, client CPU 1.071 ms per search; exact 3504/3494 MHz, performance, outside load 0.11 before the warm-up, 0.11 during, client CPU 1.132 ms per search; default@1 3497/3493 MHz, performance, outside load 0.11 before the warm-up, 0.12 during, client CPU 1.004 ms per search.

### 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
- Load: 524 rows in batches of 1000, 1.4 s of upserts (375 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: 9907 latency samples, target held 524 rows, 0 errors
- Client CPU per search: 1 searcher 0.47 ms, 8 searchers 0.51 ms
- Exact mode: 23,010 searches, p50 2.55 ms, p95 2.93 ms, recall 1.000
- RAM: 2.6 GiB (docker stats); disk: 9.44 MiB added by this load (10.46 MiB (whole engine data folder), 1.02 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 2.11 2.47 2.57 (1/5/15 min) when searching began.
- Rehearsal before any timed pass, untimed, every pass type for at least 5 s through the same code the timed passes use: default@1 1,032 searches in 5.0 s, default@8 3,754 searches in 5.0 s, exact 1,124 searches in 5.0 s.
- Settle: settled after the one extension. First settle 1.3 s and 500 searches (0 failed); p50 of the last windows of 100: 2.40, 2.43, 2.33 ms, within 5% across 3 windows; trial right before the first timed pass of 649 searches in 2.0 s: p50 2.957 ms against the settled p50 2.404 ms, 19% apart (limit 15%), so the settle was extended once (30.0 s and 14,288 searches (0 failed); p50 of the last windows of 100: 1.97, 1.98, 1.98 ms); second trial of 975 searches in 2.0 s: p50 1.969 ms against the settled p50 1.980 ms, 1% apart (limit 15%).
- Pass default@1 after settle: 2026-10-05T04:13:47.907Z to 2026-10-05T04:14:07.907Z (20.0 s), 9,907 searches, 0 failed, p50 1.971 ms, mean 2.017 ms, p99 2.677 ms, 495.3 QPS (1000/QPS 2.019 ms).
- Pass default@8 after default@1: 2026-10-05T04:14:07.931Z to 2026-10-05T04:14:27.935Z (20.0 s), 28,929 searches, 0 failed, p50 5.281 ms, mean 5.529 ms, p99 11.045 ms, 1446.2 QPS (1000/QPS 0.691 ms).
- Pass exact after default@8: 2026-10-05T04:14:27.998Z to 2026-10-05T04:15:28.000Z (60.0 s), 23,010 searches, 0 failed, p50 2.548 ms, mean 2.606 ms, p99 3.663 ms, 383.5 QPS (1000/QPS 2.608 ms).
- Passes in the order run: default@1, default@8, exact; each after its own warm-up. Untimed searches (rehearsal, settle, settle check and per-pass warm-ups): 22,402 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 (0d3ef5251680) on network engines_default at container-ip:9200, not through docker-proxy localhost:9201. 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-opensearch: already on CPUs 2-3,6-7 (its cpuset since creation), not changed. Read back: container gvb-opensearch: 55 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 0.12 before the warm-up, 0.13 during, client CPU 0.466 ms per search; default@8 3492/3492 MHz, performance, outside load 0.12 before the warm-up, 0.09 during, client CPU 0.511 ms per search; exact 3492/3492 MHz, performance, outside load 0.11 before the warm-up, 0.12 during, client CPU 0.469 ms per search.

### 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)
- Load: 524 rows in batches of 1000, 0.1 s of upserts (7,760 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: 22516 latency samples, target held 524 rows, 0 errors
- Client CPU per search: 1 searcher 0.73 ms, 8 searchers 0.53 ms
- RAM: 37.21 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 1.97 2.39 2.54 (1/5/15 min) when searching began.
- Rehearsal before any timed pass, untimed, every pass type for at least 5 s through the same code the timed passes use: default@8 16,436 searches in 5.0 s, default@1 5,638 searches in 5.0 s.
- Settle: settled after 0.3 s and 300 searches (0 failed); p50 of the last windows of 100: 0.87, 0.85, 0.87 ms, within 5% across 3 windows; confirmed by a trial right before the first timed pass of 2,266 searches in 2.0 s: p50 0.873 ms against the settled p50 0.866 ms, 1% apart (limit 15%).
- Pass default@8 after settle: 2026-10-05T04:15:49.738Z to 2026-10-05T04:16:09.740Z (20.0 s), 66,050 searches, 0 failed, p50 2.337 ms, mean 2.420 ms, p99 4.174 ms, 3302.3 QPS (1000/QPS 0.303 ms).
- Pass default@1 after default@8: 2026-10-05T04:16:09.779Z to 2026-10-05T04:16:29.779Z (20.0 s), 22,516 searches, 0 failed, p50 0.877 ms, mean 0.887 ms, p99 1.390 ms, 1125.8 QPS (1000/QPS 0.888 ms).
- Passes in the order run: default@8, default@1; each after its own warm-up. Untimed searches (rehearsal, settle, settle check and per-pass warm-ups): 24,680 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 226,492, 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 (f16908f64dfa) on network gvb-qdrant_default at container-ip:6334, not through docker-proxy localhost:16334; container gvb-qdrant (f16908f64dfa) 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 0.12 before the warm-up, 0.08 during, client CPU 0.525 ms per search; default@1 3492/3492 MHz, performance, outside load 0.09 before the warm-up, 0.09 during, client CPU 0.733 ms per search.

### 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)
- Load: 524 rows in batches of 1000, 0.9 s of upserts (560 rows/s); count matched 0.2 s after the last upsert; index step 0.2 s (FAILED after 0.2 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: 15007 latency samples, target held 524 rows, 0 errors
- Client CPU per search: 1 searcher 0.47 ms, 8 searchers 0.50 ms
- Exact mode: 46,858 searches, p50 1.26 ms, p95 1.48 ms, recall 1.000
- RAM: 2.54 GiB (docker stats); disk: 2.38 MiB added by this load (2.59 MiB (whole engine data folder), 210.04 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 2.82 2.72 2.65 (1/5/15 min) when searching began.
- Rehearsal before any timed pass, untimed, every pass type for at least 5 s through the same code the timed passes use: default@8 2,315 searches in 5.0 s, default@1 1,495 searches in 5.0 s, exact 2,195 searches in 5.0 s.
- Settle: settled after 4.1 s and 2,100 searches (0 failed); p50 of the last windows of 100: 1.59, 1.60, 1.60 ms, within 5% across 3 windows; confirmed by a trial right before the first timed pass of 1,101 searches in 2.0 s: p50 1.609 ms against the settled p50 1.596 ms, 1% apart (limit 15%).
- Pass default@8 after settle: 2026-10-05T04:17:25.477Z to 2026-10-05T04:17:45.479Z (20.0 s), 46,698 searches, 0 failed, p50 3.027 ms, mean 3.424 ms, p99 8.722 ms, 2334.7 QPS (1000/QPS 0.428 ms).
- Pass default@1 after default@8: 2026-10-05T04:17:45.519Z to 2026-10-05T04:18:05.520Z (20.0 s), 15,007 searches, 0 failed, p50 1.314 ms, mean 1.331 ms, p99 1.737 ms, 750.3 QPS (1000/QPS 1.333 ms).
- Pass exact after default@1: 2026-10-05T04:18:05.563Z to 2026-10-05T04:19:05.563Z (60.0 s), 46,858 searches, 0 failed, p50 1.257 ms, mean 1.279 ms, p99 1.693 ms, 781.0 QPS (1000/QPS 1.280 ms).
- Passes in the order run: default@8, default@1, exact; each after its own warm-up. Untimed searches (rehearsal, settle, settle check and per-pass warm-ups): 9,266 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 (1aa3acdd0424) on network engines_default at container-ip:9200, not through docker-proxy localhost:9200. 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-elasticsearch: already on CPUs 2-3,6-7 (its cpuset since creation), not changed. Read back: container gvb-elasticsearch: 75 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, 0.12 during, client CPU 0.504 ms per search; default@1 3492/3492 MHz, performance, outside load 0.13 before the warm-up, 0.11 during, client CPU 0.466 ms per search; exact 3492/3492 MHz, performance, outside load 0.12 before the warm-up, 0.12 during, client CPU 0.458 ms per search.

### 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
- Load: 524 rows in batches of 1000, 0.3 s of upserts (1,603 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: 5656 latency samples, target held 524 rows, 0 errors
- Client CPU per search: 1 searcher 4.82 ms, 8 searchers 4.94 ms
- Exact mode: 14,592 searches, p50 4.05 ms, p95 4.80 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 1.82 2.69 2.67 (1/5/15 min) when searching began.
- Rehearsal before any timed pass, untimed, every pass type for at least 5 s through the same code the timed passes use: default@1 1,430 searches in 5.0 s, exact 1,253 searches in 5.0 s, default@8 1,432 searches in 5.0 s.
- Settle: settled after 1.0 s and 300 searches (0 failed); p50 of the last windows of 100: 3.45, 3.43, 3.46 ms, within 5% across 3 windows; confirmed by a trial right before the first timed pass of 571 searches in 2.0 s: p50 3.483 ms against the settled p50 3.446 ms, 1% apart (limit 15%).
- Pass default@1 after settle: 2026-10-05T04:19:28.471Z to 2026-10-05T04:19:48.472Z (20.0 s), 5,656 searches, 0 failed, p50 3.501 ms, mean 3.534 ms, p99 4.970 ms, 282.8 QPS (1000/QPS 3.536 ms).
- Pass exact after default@1: 2026-10-05T04:19:48.569Z to 2026-10-05T04:20:48.573Z (60.0 s), 14,592 searches, 0 failed, p50 4.045 ms, mean 4.109 ms, p99 5.551 ms, 243.2 QPS (1000/QPS 4.112 ms).
- Pass default@8 after exact: 2026-10-05T04:20:48.658Z to 2026-10-05T04:21:08.686Z (20.0 s), 5,592 searches, 0 failed, p50 28.447 ms, mean 28.631 ms, p99 31.529 ms, 279.2 QPS (1000/QPS 3.582 ms).
- Passes in the order run: default@1, exact, default@8; each after its own warm-up. Untimed searches (rehearsal, settle, settle check and per-pass warm-ups): 5,046 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 3594/3592 MHz, performance, outside load 0.10 before the warm-up, 0.11 during, client CPU 4.817 ms per search; exact 3594/3592 MHz, performance, outside load 0.11 before the warm-up, 0.12 during, client CPU 6.075 ms per search; default@8 3597/3592 MHz, performance, outside load 0.12 before the warm-up, 0.09 during, client CPU 4.945 ms per search.

### 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)
- Load: 524 rows in batches of 1000, 0.7 s of upserts (707 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: 9719 latency samples, target held 524 rows, 0 errors
- Client CPU per search: 1 searcher 0.45 ms, 8 searchers 0.46 ms
- RAM: 42.68 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 1.40 2.24 2.50 (1/5/15 min) when searching began.
- Rehearsal before any timed pass, untimed, every pass type for at least 5 s through the same code the timed passes use: default@1 2,429 searches in 5.0 s, default@8 4,027 searches in 5.0 s.
- Settle: settled after 0.6 s and 300 searches (0 failed); p50 of the last windows of 100: 2.02, 2.02, 2.02 ms, within 5% across 3 windows; confirmed by a trial right before the first timed pass of 973 searches in 2.0 s: p50 2.044 ms against the settled p50 2.024 ms, 1% apart (limit 15%).
- Pass default@1 after settle: 2026-10-05T04:21:28.597Z to 2026-10-05T04:21:48.598Z (20.0 s), 9,719 searches, 0 failed, p50 2.040 ms, mean 2.056 ms, p99 2.517 ms, 485.9 QPS (1000/QPS 2.058 ms).
- Pass default@8 after default@1: 2026-10-05T04:21:48.632Z to 2026-10-05T04:22:08.640Z (20.0 s), 16,126 searches, 0 failed, p50 9.862 ms, mean 9.923 ms, p99 11.710 ms, 806.0 QPS (1000/QPS 1.241 ms).
- Passes in the order run: default@1, default@8; each after its own warm-up. Untimed searches (rehearsal, settle, settle check and per-pass warm-ups): 7,769 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 (8906b990c0ae) 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@1 3592/3592 MHz, performance, outside load 0.06 before the warm-up, 0.05 during, client CPU 0.455 ms per search; default@8 3592/3592 MHz, performance, outside load 0.05 before the warm-up, 0.05 during, client CPU 0.455 ms per search.

### 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
- Load: 524 rows in batches of 1000, 2.0 s of upserts (262 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: 10157 latency samples, target held 524 rows, 0 errors
- Client CPU per search: 1 searcher 0.95 ms, 8 searchers 0.93 ms
- Exact mode: 30,863 searches, p50 1.88 ms, p95 2.33 ms, recall 1.000
- RAM: 2.83 GiB (docker stats); disk: 6.39 MiB added by this load (1.82 GiB (whole engine data folder), 1.81 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 4.16 2.71 2.62 (1/5/15 min) when searching began.
- Rehearsal before any timed pass, untimed, every pass type for at least 5 s through the same code the timed passes use: default@8 1,673 searches in 5.0 s, default@1 1,099 searches in 5.0 s, exact 1,520 searches in 5.0 s.
- Settle: settled after 0.8 s and 300 searches (0 failed); p50 of the last windows of 100: 2.69, 2.60, 2.59 ms, within 5% across 3 windows; confirmed by a trial right before the first timed pass of 682 searches in 2.0 s: p50 2.764 ms against the settled p50 2.598 ms, 6% apart (limit 15%).
- Pass default@8 after settle: 2026-10-05T04:23:23.576Z to 2026-10-05T04:23:43.578Z (20.0 s), 19,548 searches, 0 failed, p50 7.550 ms, mean 8.183 ms, p99 17.615 ms, 977.3 QPS (1000/QPS 1.023 ms).
- Pass default@1 after default@8: 2026-10-05T04:23:43.625Z to 2026-10-05T04:24:03.627Z (20.0 s), 10,157 searches, 0 failed, p50 1.895 ms, mean 1.967 ms, p99 2.892 ms, 507.8 QPS (1000/QPS 1.969 ms).
- Pass exact after default@1: 2026-10-05T04:24:03.678Z to 2026-10-05T04:25:03.679Z (60.0 s), 30,863 searches, 0 failed, p50 1.884 ms, mean 1.942 ms, p99 2.690 ms, 514.4 QPS (1000/QPS 1.944 ms).
- Passes in the order run: default@8, default@1, exact; each after its own warm-up. Untimed searches (rehearsal, settle, settle check and per-pass warm-ups): 5,334 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 (9f3ee919f559) on network engines_default at container-ip:8080, not through docker-proxy localhost:8090; container gvb-vespa (9f3ee919f559) 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): default@8 3507/3498 MHz, performance, outside load 0.13 before the warm-up, 0.15 during, client CPU 0.928 ms per search; default@1 3515/3497 MHz, performance, outside load 0.14 before the warm-up, 0.07 during, client CPU 0.947 ms per search; exact 3521/3498 MHz, performance, outside load 0.12 before the warm-up, 0.09 during, client CPU 0.943 ms per search.

### 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
- Load: 524 rows in batches of 1000, 1.1 s of upserts (494 rows/s); count matched 0.0 s after the last upsert; index step 0.1 s (nothing to build, Typesense inserts into the HNSW graph while writing; ready after 0.1 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: 4893 latency samples, target held 524 rows, 0 errors
- Client CPU per search: 1 searcher 0.51 ms, 8 searchers 0.57 ms
- Exact mode: 10,833 searches, p50 5.48 ms, p95 5.83 ms, recall 1.000
- RAM: 118.3 MiB (docker stats); disk: 0 B added by this load (252.92 MiB (whole engine data folder), 515.88 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 5.85 3.84 3.06 (1/5/15 min) when searching began.
- Rehearsal before any timed pass, untimed, every pass type for at least 5 s through the same code the timed passes use: default@8 2,932 searches in 5.0 s, default@1 1,224 searches in 5.0 s, exact 907 searches in 5.0 s.
- Settle: settled after 1.2 s and 300 searches (0 failed); p50 of the last windows of 100: 4.06, 4.07, 4.03 ms, within 5% across 3 windows; confirmed by a trial right before the first timed pass of 489 searches in 2.0 s: p50 4.052 ms against the settled p50 4.058 ms, 0% apart (limit 15%).
- Pass default@8 after settle: 2026-10-05T04:25:40.218Z to 2026-10-05T04:26:00.224Z (20.0 s), 11,764 searches, 0 failed, p50 13.089 ms, mean 13.601 ms, p99 25.201 ms, 588.0 QPS (1000/QPS 1.701 ms).
- Pass default@1 after default@8: 2026-10-05T04:26:00.309Z to 2026-10-05T04:26:20.310Z (20.0 s), 4,893 searches, 0 failed, p50 4.049 ms, mean 4.086 ms, p99 4.649 ms, 244.6 QPS (1000/QPS 4.088 ms).
- Pass exact after default@1: 2026-10-05T04:26:20.425Z to 2026-10-05T04:27:20.427Z (60.0 s), 10,833 searches, 0 failed, p50 5.479 ms, mean 5.537 ms, p99 6.578 ms, 180.5 QPS (1000/QPS 5.539 ms).
- Passes in the order run: default@8, default@1, exact; each after its own warm-up. Untimed searches (rehearsal, settle, settle check and per-pass warm-ups): 5,912 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 (3e379dbf8d54) 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 0.08 before the warm-up, -0.05 during, client CPU 0.575 ms per search; default@1 3592/3592 MHz, performance, outside load 0.02 before the warm-up, 0.12 during, client CPU 0.513 ms per search; exact 3592/3592 MHz, performance, outside load 0.05 before the warm-up, 0.13 during, client CPU 0.522 ms per search.

### 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
- Load: 524 rows in batches of 1000, 0.2 s of upserts (3,315 rows/s); count matched 1.2 s after the last upsert; index step 1.1 s (index status READY, queryable True; mongot holds 524 of 524 documents in 2 segment(s), 2 searched through the HNSW graph (Approximate); wait took 1.1 s)
- Search: 14185 latency samples, target held 524 rows, 0 errors
- Client CPU per search: 1 searcher 0.31 ms, 8 searchers 0.28 ms
- Exact mode: 47,615 searches, p50 1.23 ms, p95 1.43 ms, recall 1.000
- RAM: 680.4 MiB (docker stats); disk: 0 B added by this load (1.3 GiB (whole engine data folder), 1.3 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), 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 2.65 3.35 2.99 (1/5/15 min) when searching began.
- Rehearsal before any timed pass, untimed, every pass type for at least 5 s through the same code the timed passes use: default@8 3,609 searches in 5.0 s, default@1 2,605 searches in 5.0 s, exact 3,108 searches in 5.0 s.
- Settle: settled after 0.7 s and 400 searches (0 failed); p50 of the last windows of 100: 1.56, 1.54, 1.53 ms, within 5% across 3 windows; confirmed by a trial right before the first timed pass of 1,230 searches in 2.0 s: p50 1.500 ms against the settled p50 1.538 ms, 2% apart (limit 15%).
- Pass default@8 after settle: 2026-10-05T04:27:51.098Z to 2026-10-05T04:28:11.100Z (20.0 s), 40,768 searches, 0 failed, p50 3.708 ms, mean 3.922 ms, p99 8.452 ms, 2038.1 QPS (1000/QPS 0.491 ms).
- Pass default@1 after default@8: 2026-10-05T04:28:11.141Z to 2026-10-05T04:28:31.141Z (20.0 s), 14,185 searches, 0 failed, p50 1.383 ms, mean 1.408 ms, p99 1.945 ms, 709.2 QPS (1000/QPS 1.410 ms).
- Pass exact after default@1: 2026-10-05T04:28:31.180Z to 2026-10-05T04:29:31.181Z (60.0 s), 47,615 searches, 0 failed, p50 1.234 ms, mean 1.258 ms, p99 1.764 ms, 793.6 QPS (1000/QPS 1.260 ms).
- Passes in the order run: default@8, default@1, exact; each after its own warm-up. Untimed searches (rehearsal, settle, settle check and per-pass warm-ups): 11,012 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), 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 (f9af1de6b95e) 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: 212 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 0.21 before the warm-up, 0.12 during, client CPU 0.281 ms per search; default@1 3492/3492 MHz, performance, outside load 0.17 before the warm-up, 0.20 during, client CPU 0.310 ms per search; exact 3492/3492 MHz, performance, outside load 0.18 before the warm-up, 0.21 during, client CPU 0.301 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)
- Load: 524 rows in batches of 1000, 0.8 s of upserts (623 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: 23525 latency samples, target held 524 rows, 0 errors
- Client CPU per search: 1 searcher 0.69 ms, 8 searchers 0.54 ms
- Exact mode: 26,870 searches, p50 2.20 ms, p95 2.42 ms, recall 1.000
- RAM: 100.4 MiB (docker stats); disk: 7.59 MiB added by this load (454.64 MiB (whole engine data folder), 447.04 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 0.76 2.48 2.75 (1/5/15 min) when searching began.
- Rehearsal before any timed pass, untimed, every pass type for at least 5 s through the same code the timed passes use: default@1 6,213 searches in 5.0 s, default@8 20,720 searches in 5.0 s, exact 2,247 searches in 5.0 s.
- Settle: settled after 0.6 s and 700 searches (0 failed); p50 of the last windows of 100: 0.83, 0.86, 0.86 ms, within 5% across 3 windows; confirmed by a trial right before the first timed pass of 2,388 searches in 2.0 s: p50 0.821 ms against the settled p50 0.860 ms, 5% apart (limit 15%).
- Pass default@1 after settle: 2026-10-05T04:30:58.034Z to 2026-10-05T04:31:18.035Z (20.0 s), 23,525 searches, 0 failed, p50 0.834 ms, mean 0.848 ms, p99 1.244 ms, 1176.2 QPS (1000/QPS 0.850 ms).
- Pass default@8 after default@1: 2026-10-05T04:31:18.063Z to 2026-10-05T04:31:38.065Z (20.0 s), 82,415 searches, 0 failed, p50 1.896 ms, mean 1.940 ms, p99 3.419 ms, 4120.5 QPS (1000/QPS 0.243 ms).
- Pass exact after default@8: 2026-10-05T04:31:38.137Z to 2026-10-05T04:32:38.137Z (60.0 s), 26,870 searches, 0 failed, p50 2.196 ms, mean 2.231 ms, p99 3.016 ms, 447.8 QPS (1000/QPS 2.233 ms).
- Passes in the order run: default@1, default@8, exact; each after its own warm-up. Untimed searches (rehearsal, settle, settle check and per-pass warm-ups): 32,328 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 (761f565917c8) 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@1 3492/3492 MHz, performance, outside load 0.11 before the warm-up, 0.11 during, client CPU 0.686 ms per search; default@8 3492/3492 MHz, performance, outside load 0.11 before the warm-up, 0.10 during, client CPU 0.542 ms per search; exact 3492/3492 MHz, performance, outside load 0.11 before the warm-up, 0.11 during, client CPU 0.688 ms per search.

### 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
- Load: 524 rows in batches of 1000, 0.2 s of upserts (2,882 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: 3533 latency samples, target held 524 rows, 0 errors
- Client CPU per search: 1 searcher 0.90 ms, 8 searchers 1.04 ms
- Exact mode: 9,008 searches, p50 6.37 ms, p95 9.30 ms, recall 1.000
- RAM: 738.5 MiB (docker stats); disk: 2.42 GiB added by this load (5.95 GiB (whole engine data folder), 3.53 GiB 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 2.57 2.81 2.85 (1/5/15 min) when searching began.
- Rehearsal before any timed pass, untimed, every pass type for at least 5 s through the same code the timed passes use: exact 729 searches in 5.0 s, default@8 2,413 searches in 5.0 s, default@1 981 searches in 5.0 s.
- Settle: settled after 1.5 s and 300 searches (0 failed); p50 of the last windows of 100: 4.75, 4.69, 4.68 ms, within 5% across 3 windows; confirmed by a trial right before the first timed pass of 406 searches in 2.0 s: p50 4.724 ms against the settled p50 4.690 ms, 1% apart (limit 15%).
- Pass exact after settle: 2026-10-05T04:33:11.162Z to 2026-10-05T04:34:11.167Z (60.0 s), 9,008 searches, 0 failed, p50 6.367 ms, mean 6.658 ms, p99 10.070 ms, 150.1 QPS (1000/QPS 6.661 ms).
- Pass default@8 after exact: 2026-10-05T04:34:11.230Z to 2026-10-05T04:34:31.244Z (20.0 s), 8,268 searches, 0 failed, p50 18.696 ms, mean 19.355 ms, p99 36.642 ms, 413.1 QPS (1000/QPS 2.421 ms).
- Pass default@1 after default@8: 2026-10-05T04:34:31.368Z to 2026-10-05T04:34:51.370Z (20.0 s), 3,533 searches, 0 failed, p50 5.200 ms, mean 5.658 ms, p99 9.063 ms, 176.6 QPS (1000/QPS 5.661 ms).
- Passes in the order run: exact, default@8, default@1; each after its own warm-up. Untimed searches (rehearsal, settle, settle check and per-pass warm-ups): 4,889 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 (5e4ed8528b2f) on network gvb-clickhouse_default at container-ip:8123, not through docker-proxy localhost:8123. Open after the passes: container-ip:8123 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-clickhouse: already on CPUs 2-3,6-7 (its cpuset since creation), not changed. Read back: container gvb-clickhouse: 724 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 3592/3592 MHz, performance, outside load 0.13 before the warm-up, 0.13 during, client CPU 0.860 ms per search; default@8 3554/3564 MHz, performance, outside load 0.13 before the warm-up, 0.13 during, client CPU 1.044 ms per search; default@1 3592/3591 MHz, performance, outside load 0.13 before the warm-up, 0.19 during, client CPU 0.903 ms per search.

### 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)
- Load: 524 rows in batches of 1000, 0.4 s of upserts (1,188 rows/s); count matched 1.7 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: 8523 latency samples, target held 524 rows, 0 errors
- Client CPU per search: 1 searcher 0.55 ms, 8 searchers 0.64 ms
- RAM: 198.4 MiB (docker stats); disk: 10.86 MiB added by this load (166.8 MiB (whole engine data folder), 155.93 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 0.43 2.01 2.57 (1/5/15 min) when searching began.
- Rehearsal before any timed pass, untimed, every pass type for at least 5 s through the same code the timed passes use: default@1 2,094 searches in 5.0 s, default@8 6,293 searches in 5.0 s.
- Settle: settled after 0.7 s and 300 searches (0 failed); p50 of the last windows of 100: 2.25, 2.24, 2.26 ms, within 5% across 3 windows; confirmed by a trial right before the first timed pass of 848 searches in 2.0 s: p50 2.239 ms against the settled p50 2.248 ms, 0% apart (limit 15%).
- Pass default@1 after settle: 2026-10-05T04:37:41.212Z to 2026-10-05T04:38:01.213Z (20.0 s), 8,523 searches, 0 failed, p50 2.239 ms, mean 2.345 ms, p99 3.716 ms, 426.1 QPS (1000/QPS 2.347 ms).
- Pass default@8 after default@1: 2026-10-05T04:38:01.246Z to 2026-10-05T04:38:21.249Z (20.0 s), 25,214 searches, 0 failed, p50 5.780 ms, mean 6.344 ms, p99 15.219 ms, 1260.5 QPS (1000/QPS 0.793 ms).
- Passes in the order run: default@1, default@8; each after its own warm-up. Untimed searches (rehearsal, settle, settle check and per-pass warm-ups): 9,575 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 43312 searches (searchParams.params.ef=100, Strong consistency), the query node counted 43312 search requests for this collection, segment searches Sealed +42537 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 (9e95ecdb6c14) on network gvb-milvus_default at container-ip:19530, not through docker-proxy localhost:19530; container gvb-milvus (9e95ecdb6c14) 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 3538/3501 MHz, performance, outside load 0.14 before the warm-up, 0.10 during, client CPU 0.553 ms per search; default@8 3492/3493 MHz, performance, outside load 0.12 before the warm-up, 0.14 during, client CPU 0.635 ms per search.


## Notes

- Order: targets ran one at a time in a random order from seed 503 (runSeed; --seed 503 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 for 5 s each through the same code the passes use, then a settle (default searches one at a time until the p50s of 3 windows of 100 agree within 5%, at most 120 s). Right before the first timed pass a 2 s trial of default searches must agree with the settled p50 within 15%; if it does not, or the settle did not settle, the settle is extended once (at least 30 s, at most 120 s more) and a second trial taken. Each target's notes give the result; a target still disagreeing is marked NOT settled.
- Warm-up: every timed pass started with its own untimed warm-up of 20 searches in the same search mode and with the same number of searchers, stopped early after 60 s; with machine control on, the check for a quiet box comes right before the warm-up, so warm-up and timed pass run back to back. 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.
- 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.
- sqlitevec is embedded: it ran inside the client process on the client CPUs 0-1,4-5, sharing them with the client.
- duckdb is embedded: it ran inside the client process on the client CPUs 0-1,4-5, sharing them with the client.
- WARNING: accounting mismatch in typesense default@8: outside load during the pass was -0.051 CPUs, below -0.05; the kernel's busy time minus this process minus the engine's cgroups came out negative, so the outside-load figures of this pass cannot be trusted.