{ "startedUtc": "2026-10-03T15:15:06Z", "command": "run-all", "commandLine": "~/tmp/gvblane-bench/src/GenericVectorBuilder.Bench/bin/Debug/net10.0/GenericVectorBuilder.Bench.dll run-all --pipeline adventureworks --limit 50000 --targets sql,qdrant,qdrant-hnsw,sql-diskann --queries random:200 --repo ~/ForClaude/GenericVectorBuilder", "machine": { "host": "bench-host", "cpu": "Intel(R) Xeon(R) CPU E5-1620 v3 @ 3.50GHz", "logicalCpus": 8, "ramGiB": 62.7, "gpu": "NVIDIA GeForce RTX 3060, 12288 MiB", "os": "Ubuntu 24.04.5 LTS, kernel 6.8.0-139-generic", "dotNet": ".NET 10.0.12", "loadAverage": "14.58 19.60 23.60" }, "pipeline": "adventureworks", "collection": "gvbbench_adventureworks", "rows": 50000, "dimension": 1024, "source": "Read READ-ONLY from GenericVectorBuilder.dbo.gvb_adventureworks in ChunkId order, vectors as native SqlVector\u003Cfloat\u003E, 2.2 s.", "queries": "random:200 (stored vectors picked with seed 20261003; each query\u0027s own row is left out of the exact answer and removed from every engine\u0027s hits)", "queryCount": 200, "top": 10, "concurrency": [ 1, 8 ], "secondsPerLevel": 20, "truthSeconds": 1.2339391, "notes": [ "WARNING: load average 14.58 19.60 23.60 on 8 logical CPUs when the run started. Other work was competing for the CPU, so absolute latency and QPS are worse than this box can do; compare engines only within one run, and rerun on a quiet box before quoting numbers.", "Load rows/s counts only time inside each target\u0027s upsert calls: one writer, batches of 1000, rows already in memory, the collection dropped and created fresh first.", "Latency is client-side wall time around each search (network and driver included), one query at a time, after 20 warm-up queries; at least 200 samples (small query sets are repeated).", "QPS: N workers searching back to back for 20 s per level; completed searches divided by 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.", "RAM of always-on servers (SQL Server, Qdrant) is the whole process, including every other database or collection it serves; for compose engines it is docker stats of the engine\u0027s containers. Disk of compose engines is the whole engine data folder.", "The empty benchmark database GvbBench was dropped at the end." ], "targets": [ { "name": "sql", "engine": "Microsoft SQL Server 2025 (RTM-CU9) (KB5122048) - 17.0.5005.3 (X64)", "index": "exact VECTOR_DISTANCE cosine, no vector index (full scan)", "hosting": "always-on", "load": { "rows": 50000, "batch": 1000, "upsertSeconds": 137.3601409, "rowsPerSecond": 364.0066155465046, "countMatchSeconds": 0.0213642 }, "search": { "countInTarget": 50000, "latencySamples": 200, "p50Ms": 66.76554999999999, "p95Ms": 79.904885, "p99Ms": 87.49519699999998, "qps": { "1": 14.118282431420866, "8": 11.86263017857104 }, "recall": 1, "errors": 0, "exactQueries": 0 }, "ram": { "bytes": 4431282176, "text": "4.13 GiB (whole SQL Server process)" }, "disk": { "bytes": 418004992, "text": "398.64 MiB (table and its indexes, reserved pages)" }, "notes": [ "WARNING: load average 8.02 15.31 21.42 (1/5/15 min) on 8 logical CPUs when searching began; the box was busy, so latency and QPS are inflated by other work.", "Benchmark copy gvbbench_adventureworks dropped afterwards." ] }, { "name": "qdrant", "engine": "Qdrant 1.17.0 (systemd, local)", "index": "exact search (the builder\u0027s setting), HNSW m=16 ef_construct=100 built but not used", "hosting": "always-on", "load": { "rows": 50000, "batch": 1000, "upsertSeconds": 18.1513944, "rowsPerSecond": 2754.6093097949542, "countMatchSeconds": 0.0425281, "indexSeconds": 7.0340484, "indexNote": "status Green, 50,000 of 50,000 vectors in HNSW segments (4 segments)" }, "search": { "countInTarget": 50000, "latencySamples": 200, "p50Ms": 12.2446, "p95Ms": 14.520425, "p99Ms": 16.178955999999992, "qps": { "1": 88.40507033312996, "8": 124.23719868707566 }, "recall": 1, "errors": 0, "exactQueries": 0 }, "ram": { "bytes": 6650314752, "text": "6.19 GiB (whole Qdrant process, every collection)" }, "disk": { "bytes": 481401374, "text": "459.1 MiB (collection folder)" }, "notes": [ "WARNING: load average 9.12 13.81 20.38 (1/5/15 min) on 8 logical CPUs when searching began; the box was busy, so latency and QPS are inflated by other work.", "Benchmark copy gvbbench_adventureworks dropped afterwards." ] }, { "name": "qdrant-hnsw", "engine": "Qdrant 1.17.0 (systemd, local)", "index": "HNSW m=16 ef_construct=100, hnsw_ef=server default, cosine", "hosting": "always-on", "load": { "rows": 50000, "batch": 1000, "upsertSeconds": 16.1069196, "rowsPerSecond": 3104.255887637261, "countMatchSeconds": 0.016156, "indexSeconds": 4.0059458, "indexNote": "status Green, 50,000 of 50,000 vectors in HNSW segments (4 segments)" }, "search": { "countInTarget": 50000, "latencySamples": 200, "p50Ms": 1.60535, "p95Ms": 2.1378449999999987, "p99Ms": 2.8832809999999967, "qps": { "1": 462.33725105030226, "8": 1466.646895419358 }, "recall": 0.9984999999999999, "errors": 0, "exactQueries": 200, "exactP50Ms": 10.8719, "exactP95Ms": 13.43191, "exactRecall": 1 }, "ram": { "bytes": 6688403456, "text": "6.23 GiB (whole Qdrant process, every collection)" }, "disk": { "bytes": 514946680, "text": "491.09 MiB (collection folder)" }, "notes": [ "WARNING: load average 8.99 12.93 19.63 (1/5/15 min) on 8 logical CPUs when searching began; the box was busy, so latency and QPS are inflated by other work.", "Benchmark copy gvbbench_adventureworks dropped afterwards." ] }, { "name": "sql-diskann", "engine": "Microsoft SQL Server 2025 (RTM-CU9) (KB5122048) - 17.0.5005.3 (X64)", "index": "DiskANN (preview) via VECTOR_SEARCH, cosine, build {\u0022StartId\u0022:\u002230588\u0022, \u0022L\u0022:\u002248\u0022, \u0022M\u0022:\u00228\u0022, \u0022R\u0022:\u002248\u0022}", "hosting": "always-on", "load": { "rows": 50000, "batch": 1000, "upsertSeconds": 71.5365479, "rowsPerSecond": 698.9434277691781, "countMatchSeconds": 0.0122078, "indexSeconds": 126.7477653, "indexNote": "copied to dbo.gvb_gvbbench_adventureworks_ann (INT key) and built DiskANN, parameters {\u0022StartId\u0022:\u002230588\u0022, \u0022L\u0022:\u002248\u0022, \u0022M\u0022:\u00228\u0022, \u0022R\u0022:\u002248\u0022}" }, "search": { "countInTarget": 50000, "latencySamples": 200, "p50Ms": 6.8722, "p95Ms": 10.268509999999997, "p99Ms": 11.253286999999991, "qps": { "1": 146.7150502742989, "8": 471.40446898233057 }, "recall": 0.9975, "errors": 0, "exactQueries": 200, "exactP50Ms": 86.637, "exactP95Ms": 130.53921999999994, "exactRecall": 1 }, "ram": { "bytes": 4431282176, "text": "4.13 GiB (whole SQL Server process)" }, "disk": { "bytes": 424230912, "text": "404.58 MiB (table and its indexes, reserved pages)" }, "notes": [ "WARNING: load average 8.72 10.19 16.95 (1/5/15 min) on 8 logical CPUs when searching began; the box was busy, so latency and QPS are inflated by other work.", "Benchmark copy gvbbench_adventureworks dropped afterwards." ] } ] }