← all runs
Cohere-50M — 2026-07-13
r8i.2xlarge · peak recall 0.9958 @ 42 QPS
QPS at recall
| Recall | Meerkat | pgvector HNSW (halfvec) | Δ vs best competitor |
|---|
| 0.90 | 634 | 227 | +179% |
| 0.95 | 332 | 26 | +1163% |
| 0.99 | 79 | — (max r=0.985) | — |
Latency at recall
| Meerkat | pgvector HNSW (halfvec) | Δ vs best competitor |
|---|
| @ recall 0.90 | | | |
| p50 | 1.54 ms | 4.28 ms | -64% |
| p95 | 1.92 ms | 6.78 ms | -72% |
| p99 | 2.16 ms | 8.65 ms | -75% |
| @ recall 0.95 | | | |
| p50 | 2.94 ms | 33.39 ms | -91% |
| p95 | 3.68 ms | 72.41 ms | -95% |
| p99 | 4.27 ms | 93.90 ms | -95% |
| @ recall 0.99 | | | |
| p50 | 12.44 ms | — | — |
| p95 | 15.45 ms | — | — |
| p99 | 18.59 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-n240000-cctrue-fs16-soar-cfs-eps035 · pgvector HNSW (halfvec) @0.90: hnsw-m24-halfvec · Meerkat @0.95: mkt-n240000-cctrue-fs16-soar-cfs-eps035 · pgvector HNSW (halfvec) @0.95: hnsw-m24-halfvec · Meerkat @0.99: mkt-n240000-cctrue-fs16-soar-cfs-eps035
Build comparison
| Meerkat | pgvector HNSW (halfvec) | Δ vs best competitor |
|---|
| Build time | 2954 s | 13866 s | -79% |
| Index size | 13.6 GB | 95.3 GB | -86% |
Build stats are for the configuration backing each column's best operating point: Meerkat = mkt-n240000-cctrue-fs16-soar-cfs-eps035, prewarmed · pgvector HNSW (halfvec) = hnsw-m24-halfvec, prewarmed.
Plots
Pareto svg ·
Distance-recall svg ·
Per-config svg ·
Build times svg ·
Index sizes svgHost
| Instance type | r8i.2xlarge |
| Hostname | ip-172-31-66-83 |
| CPU | Intel(R) Xeon(R) 6975P-C (8 vCPU) |
| Memory | 62 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) |
| pg_prewarm | 1.2 |
| vector | 0.8.5 |
Search
Index builds
| Index | Type | Build time | Size | Workers |
|---|
| hnsw-m24-halfvec | pgvector-hnsw | 13865.9s | 95.3 GB | 24 |
| mkt-n240000-ccfalse-fs16-soar-eps015 | meerkat | 5655.3s | 13.9 GB | 24 |
| mkt-n240000-cctrue-fs16-soar-cfs-eps015 | meerkat | 3422.0s | 12.9 GB | 24 |
| mkt-n240000-cctrue-fs16-soar-cfs-eps025 | meerkat | 3134.8s | 13.4 GB | 24 |
| mkt-n240000-cctrue-fs16-soar-cfs-eps035 | meerkat | 2954.2s | 13.6 GB | 24 |
Index parameters
hnsw-m24-halfvec: {"ef_construction":200,"encoding":"halfvec","m":24}
mkt-n240000-ccfalse-fs16-soar-eps015: {"boundary_epsilon":0.15,"centroid_compression":false,"centroid_fastscan":null,"fan_out":null,"fastscan":true,"fastscan_bits":null,"nlist":240000,"rerank_vectors":null,"soar_lambda":1.0}
mkt-n240000-cctrue-fs16-soar-cfs-eps015: {"boundary_epsilon":0.15,"centroid_compression":true,"centroid_fastscan":true,"fan_out":null,"fastscan":true,"fastscan_bits":null,"nlist":240000,"rerank_vectors":null,"soar_lambda":1.0}
mkt-n240000-cctrue-fs16-soar-cfs-eps025: {"boundary_epsilon":0.25,"centroid_compression":true,"centroid_fastscan":true,"fan_out":null,"fastscan":true,"fastscan_bits":null,"nlist":240000,"rerank_vectors":null,"soar_lambda":1.0}
mkt-n240000-cctrue-fs16-soar-cfs-eps035: {"boundary_epsilon":0.35,"centroid_compression":true,"centroid_fastscan":true,"fan_out":null,"fastscan":true,"fastscan_bits":null,"nlist":240000,"rerank_vectors":null,"soar_lambda":1.0}
PostgreSQL settings
effective_cache_size = 48 GB
effective_io_concurrency = 256
huge_pages = try
maintenance_work_mem = 25 GB
max_connections = 100
max_parallel_maintenance_workers = 4
max_parallel_workers = 8
random_page_cost = 1.1
shared_buffers = 16 GB
wal_level = replica
work_mem = 512 MB
