← all runs
Cohere-100M — 2026-07-15
r8i.24xlarge · peak recall 0.9969 @ 24 QPS
“100M dataset build on 24xlarge downsized to 8xlarge”
QPS at recall
| Recall | Meerkat | pgvector HNSW (halfvec) | Δ vs best competitor |
|---|
| 0.90 | 489 | 278 | +76% |
| 0.95 | 235 | 139 | +69% |
| 0.99 | 53 | — (max r=0.986) | — |
Latency at recall
| Meerkat | pgvector HNSW (halfvec) | Δ vs best competitor |
|---|
| @ recall 0.90 | | | |
| p50 | 2.01 ms | 3.63 ms | -45% |
| p95 | 2.59 ms | 4.85 ms | -47% |
| p99 | 3.00 ms | 5.79 ms | -48% |
| @ recall 0.95 | | | |
| p50 | 4.17 ms | 7.46 ms | -44% |
| p95 | 5.24 ms | 9.15 ms | -43% |
| p99 | 6.36 ms | 10.16 ms | -37% |
| @ recall 0.99 | | | |
| p50 | 18.49 ms | — | — |
| p95 | 22.39 ms | — | — |
| p99 | 28.70 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-defaults · pgvector HNSW (halfvec) @0.90: hnsw-m24-halfvec · Meerkat @0.95: mkt-defaults · pgvector HNSW (halfvec) @0.95: hnsw-m24-halfvec · Meerkat @0.99: mkt-defaults
Build comparison
| Meerkat | pgvector HNSW (halfvec) | Δ vs best competitor |
|---|
| Build time | 3443 s | 22212 s | -85% |
| Index size | 24.9 GB | 179.2 GB | -86% |
Build stats are for the configuration backing each column's best operating point: Meerkat = mkt-defaults, 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.24xlarge |
| Hostname | ip-172-31-75-2 |
| 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 (c1f5545 / 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 (1f3ad94) |
| pg_prewarm | 1.2 |
| vector | 0.8.5 |
Search
Index builds
| Index | Type | Build time | Size | Workers |
|---|
| hnsw-m24-halfvec | pgvector-hnsw | 22212.1s | 179.2 GB | 32 |
| mkt-defaults | meerkat | 3442.7s | 24.9 GB | 32 |
Index parameters
hnsw-m24-halfvec: {"ef_construction":200,"encoding":"halfvec","m":24}
mkt-defaults: {"boundary_epsilon":null,"centroid_compression":null,"centroid_fastscan":null,"fan_out":null,"fastscan":null,"fastscan_bits":null,"nlist":null,"rerank_vectors":null,"soar_lambda":null}
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
