← all runs

Cohere-5M — 2026-06-09

ip-172-31-4-132 · peak recall 0.9996 @ 14 QPS

QPS at recall

RecallMeerkatpgvector HNSWΔ vs best competitor
0.90635391+62%
0.95358295+22%
0.9912681+56%

Latency at recall

Meerkatpgvector HNSWΔ vs best competitor
@ recall 0.90
p501.55 ms2.08 ms-26%
p951.95 ms5.58 ms-65%
p992.15 ms7.02 ms-69%
@ recall 0.95
p502.76 ms3.32 ms-17%
p953.34 ms4.96 ms-33%
p993.63 ms5.97 ms-39%
@ recall 0.99
p507.94 ms12.57 ms-37%
p959.09 ms15.90 ms-43%
p999.50 ms16.78 ms-43%

All values are taken at exactly the stated recall: each configuration's own sweep is log-interpolated between its two surrounding measured points (measured values are used verbatim on an exact hit) and the best configuration at that recall is shown — values are never interpolated across different configurations. “—” means no single sweep spans that recall; nothing is extrapolated. Recall is distance-based (ann-benchmarks, tie-aware). Insert/update/delete workloads are not measured.

Interpolated on config — Meerkat @0.90: mkt-n16000-cctrue-fs16-soar · pgvector HNSW @0.90: hnsw-m24 · Meerkat @0.95: mkt-n16000-ccfalse-fs16-soar · pgvector HNSW @0.95: hnsw-m24 · Meerkat @0.99: mkt-n16000-ccfalse-fs16-soar · pgvector HNSW @0.99: hnsw-m24

Build comparison

Meerkatpgvector HNSWΔ vs best competitor
Build time830 s5015 s-83%
Index size2.2 GB19.1 GB-89%

Build stats are for the configuration backing each column's best operating point: Meerkat = mkt-n16000-ccfalse-fs16-soar · pgvector HNSW = hnsw-m24.

Plots

Pareto svg  ·  Distance-recall svg  ·  Per-config svg  ·  Build times svg  ·  Index sizes svg
paretodistance-recallbuild-timesbuild-sizes

Host

Instance type—
Hostnameip-172-31-4-132
CPUIntel(R) Xeon(R) 6975P-C (8 vCPU)
Memory62 GB
OSUbuntu 24.04.4 LTS
Kernel6.17.0-1017-aws

Versions

rekall0.1.0 (— / —)
PostgreSQLPostgreSQL 18.4 (Ubuntu 18.4-1.pgdg24.04+1) on x86_64-pc-linux-gnu, compiled by gcc (Ubuntu 13.3.0-6ubuntu2~24.04.1) 13.3.0, 64-bit
meerkat0.1.0 (65df24d)
pg_prewarm1.2
vector0.8.2

Search

S3 archives3://rekall-bench/runs/cohere-5m/2026-06-09/ip-172-31-4-132-201518/

Index builds

IndexTypeBuild timeSizeWorkers
hnsw-m24pgvector-hnsw5014.8s19.1 GB4
hnsw-m24-bqpgvector-hnsw2122.5s2.3 GB4
mkt-n16000-ccfalse-fs16meerkat701.3s2.2 GB4
mkt-n16000-ccfalse-fs16-soarmeerkat830.1s2.2 GB4
mkt-n16000-cctrue-fs16meerkat819.9s2.2 GB4
mkt-n16000-cctrue-fs16-soarmeerkat967.9s2.2 GB4
mkt-n2000-ccfalse-fs16meerkat223.0s2.1 GB4
mkt-n2000-ccfalse-fs16-soarmeerkat262.3s2.1 GB4
mkt-n2000-cctrue-fs16-soarmeerkat285.4s2.1 GB4
mkt-n4000-ccfalse-fs16meerkat205.6s2.1 GB4
mkt-n4000-ccfalse-fs16-soarmeerkat254.6s2.1 GB4
mkt-n4000-cctrue-fs16meerkat361.0s2.1 GB4
mkt-n4000-cctrue-fs16-soarmeerkat796.2s2.1 GB4
mkt-n8000-ccfalse-fs16meerkat424.2s2.1 GB4
mkt-n8000-ccfalse-fs16-soarmeerkat472.4s2.1 GB4
mkt-n8000-cctrue-fs16meerkat607.9s2.1 GB4
mkt-n8000-cctrue-fs16-soarmeerkat835.9s2.1 GB4
Index parameters
hnsw-m24: {"column_type":"vector","ef_construction":200,"m":24}
hnsw-m24-bq: {"column_type":"bit","ef_construction":200,"m":24}
mkt-n16000-ccfalse-fs16: {"boundary_epsilon":0.05,"centroid_compression":false,"fan_out":null,"fastscan":true,"fastscan_bits":null,"nlist":16000,"rerank_vectors":null,"soar_lambda":null}
mkt-n16000-ccfalse-fs16-soar: {"boundary_epsilon":0.05,"centroid_compression":false,"fan_out":null,"fastscan":true,"fastscan_bits":null,"nlist":16000,"rerank_vectors":null,"soar_lambda":1.0}
mkt-n16000-cctrue-fs16: {"boundary_epsilon":0.05,"centroid_compression":true,"fan_out":null,"fastscan":true,"fastscan_bits":null,"nlist":16000,"rerank_vectors":null,"soar_lambda":null}
mkt-n16000-cctrue-fs16-soar: {"boundary_epsilon":0.05,"centroid_compression":true,"fan_out":null,"fastscan":true,"fastscan_bits":null,"nlist":16000,"rerank_vectors":null,"soar_lambda":1.0}
mkt-n2000-ccfalse-fs16: {"boundary_epsilon":0.05,"centroid_compression":false,"fan_out":null,"fastscan":true,"fastscan_bits":null,"nlist":2000,"rerank_vectors":null,"soar_lambda":null}
mkt-n2000-ccfalse-fs16-soar: {"boundary_epsilon":0.05,"centroid_compression":false,"fan_out":null,"fastscan":true,"fastscan_bits":null,"nlist":2000,"rerank_vectors":null,"soar_lambda":1.0}
mkt-n2000-cctrue-fs16-soar: {"boundary_epsilon":0.05,"centroid_compression":true,"fan_out":null,"fastscan":true,"fastscan_bits":null,"nlist":2000,"rerank_vectors":null,"soar_lambda":1.0}
mkt-n4000-ccfalse-fs16: {"boundary_epsilon":0.05,"centroid_compression":false,"fan_out":null,"fastscan":true,"fastscan_bits":null,"nlist":4000,"rerank_vectors":null,"soar_lambda":null}
mkt-n4000-ccfalse-fs16-soar: {"boundary_epsilon":0.05,"centroid_compression":false,"fan_out":null,"fastscan":true,"fastscan_bits":null,"nlist":4000,"rerank_vectors":null,"soar_lambda":1.0}
mkt-n4000-cctrue-fs16: {"boundary_epsilon":0.05,"centroid_compression":true,"fan_out":null,"fastscan":true,"fastscan_bits":null,"nlist":4000,"rerank_vectors":null,"soar_lambda":null}
mkt-n4000-cctrue-fs16-soar: {"boundary_epsilon":0.05,"centroid_compression":true,"fan_out":null,"fastscan":true,"fastscan_bits":null,"nlist":4000,"rerank_vectors":null,"soar_lambda":1.0}
mkt-n8000-ccfalse-fs16: {"boundary_epsilon":0.05,"centroid_compression":false,"fan_out":null,"fastscan":true,"fastscan_bits":null,"nlist":8000,"rerank_vectors":null,"soar_lambda":null}
mkt-n8000-ccfalse-fs16-soar: {"boundary_epsilon":0.05,"centroid_compression":false,"fan_out":null,"fastscan":true,"fastscan_bits":null,"nlist":8000,"rerank_vectors":null,"soar_lambda":1.0}
mkt-n8000-cctrue-fs16: {"boundary_epsilon":0.05,"centroid_compression":true,"fan_out":null,"fastscan":true,"fastscan_bits":null,"nlist":8000,"rerank_vectors":null,"soar_lambda":null}
mkt-n8000-cctrue-fs16-soar: {"boundary_epsilon":0.05,"centroid_compression":true,"fan_out":null,"fastscan":true,"fastscan_bits":null,"nlist":8000,"rerank_vectors":null,"soar_lambda":1.0}
PostgreSQL settings
effective_cache_size = 48 GB
effective_io_concurrency = 256
huge_pages = try
maintenance_work_mem = 24 GB
max_connections = 100
max_parallel_maintenance_workers = 4
max_parallel_workers = 8
random_page_cost = 4
shared_buffers = 16 GB
wal_level = replica
work_mem = 512 MB