← all runs
Cohere-100M — 2026-07-13
r8i.24xlarge · peak recall 0.9924 @ 50 QPS
QPS at recall
| Recall | Meerkat | pgvector HNSW (halfvec) | Δ vs best competitor |
|---|
| 0.90 | 505 | 256 | +97% |
| 0.95 | 260 | 125 | +107% |
| 0.99 | 60 | — (max r=0.985) | — |
Latency at recall
| Meerkat | pgvector HNSW (halfvec) | Δ vs best competitor |
|---|
| @ recall 0.90 | | | |
| p50 | 1.96 ms | 3.95 ms | -50% |
| p95 | 2.35 ms | 5.27 ms | -55% |
| p99 | 2.56 ms | 6.36 ms | -60% |
| @ recall 0.95 | | | |
| p50 | 3.81 ms | 8.28 ms | -54% |
| p95 | 4.53 ms | 10.13 ms | -55% |
| p99 | 4.92 ms | 11.27 ms | -56% |
| @ recall 0.99 | | | |
| p50 | 16.51 ms | — | — |
| p95 | 19.27 ms | — | — |
| p99 | 20.59 ms | — | — |
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-n400000-eps35f · pgvector HNSW (halfvec) @0.90: hnsw-m24-halfvec · Meerkat @0.95: mkt-n400000-eps35f · pgvector HNSW (halfvec) @0.95: hnsw-m24-halfvec · Meerkat @0.99: mkt-n400000-eps35f
Build comparison
| Meerkat | pgvector HNSW (halfvec) | Δ vs best competitor |
|---|
| Build time | 3432 s | 22012 s | -84% |
| Index size | 25.1 GB | 179.2 GB | -86% |
Build stats are for the configuration backing each column's best operating point: Meerkat = mkt-n400000-eps35f, prewarmed · pgvector HNSW (halfvec) = hnsw-m24-halfvec.
Plots
Pareto svg ·
Distance-recall svg ·
Per-config svg ·
Build times svg ·
Index sizes svgHost
| Instance type | r8i.24xlarge |
| Hostname | ip-172-31-66-31 |
| CPU | Intel(R) Xeon(R) 6975P-C (96 vCPU) |
| Memory | 744 GB |
| OS | Ubuntu 26.04 LTS |
| Kernel | 7.0.0-1008-aws |
Versions
| rekall | 0.1.0 (8b64f86 / main) |
| PostgreSQL | PostgreSQL 18.4 (Ubuntu 18.4-1.pgdg26.04+1) on x86_64-pc-linux-gnu, compiled by gcc (Ubuntu 15.2.0-16ubuntu1) 15.2.0, 64-bit |
| meerkat | 0.1.0 (71e3cff) |
| pageinspect | 1.13 |
| pg_buffercache | 1.6 |
| pg_prewarm | 1.2 |
| vector | 0.8.4 |
Search
Index builds
| Index | Type | Build time | Size | Workers |
|---|
| hnsw-m24-halfvec | pgvector-hnsw | 22012.2s | 179.2 GB | 32 |
| mkt-n400000-eps15f | meerkat | 5673.0s | 23.8 GB | 32 |
| mkt-n400000-eps25f | meerkat | 3325.0s | 24.8 GB | 32 |
| mkt-n400000-eps35f | meerkat | 3431.9s | 25.1 GB | 32 |
| mkt-n400000-fixed | meerkat | 6210.0s | 20.2 GB | 32 |
Index parameters
hnsw-m24-halfvec: {"ef_construction":200,"encoding":"halfvec","m":24}
mkt-n400000-eps15f: {"boundary_epsilon":0.15,"centroid_compression":true,"centroid_fastscan":true,"fan_out":null,"fastscan":true,"fastscan_bits":null,"nlist":400000,"rerank_vectors":null,"soar_lambda":1.0}
mkt-n400000-eps25f: {"boundary_epsilon":0.25,"centroid_compression":true,"centroid_fastscan":true,"fan_out":null,"fastscan":true,"fastscan_bits":null,"nlist":400000,"rerank_vectors":null,"soar_lambda":1.0}
mkt-n400000-eps35f: {"boundary_epsilon":0.35,"centroid_compression":true,"centroid_fastscan":true,"fan_out":null,"fastscan":true,"fastscan_bits":null,"nlist":400000,"rerank_vectors":null,"soar_lambda":1.0}
mkt-n400000-fixed: {"boundary_epsilon":0.05,"centroid_compression":true,"centroid_fastscan":true,"fan_out":null,"fastscan":true,"fastscan_bits":null,"nlist":400000,"rerank_vectors":null,"soar_lambda":1.0}
PostgreSQL settings
effective_cache_size = 576 GB
effective_io_concurrency = 256
huge_pages = try
maintenance_work_mem = 307 GB
max_connections = 100
max_parallel_maintenance_workers = 32
max_parallel_workers = 96
random_page_cost = 1.1
shared_buffers = 192 GB
wal_level = replica
work_mem = 1 GB
