← zurück zur Übersicht

2026-06-11T08-02-46Z_pgvector-S0-filtered

pgvectorStufe S0 · 2.00 GBfilteredStatus: ok

Metriken

Throughput
372.8 QPS
Latenz ⌀
2.66 ms
Latenz p50
2.65 ms
Latenz p95
3.24 ms
Latenz p99
3.56 ms
Recall@1
85.00%
Recall@10
78.00%
Recall@100
46.16%
Precision@10
81.69%
NDCG@10
84.24%

Rohdaten

Konfiguration

Config-Name
pgvector-S0-filtered
DB-Version
Image
Index-Typ
hnsw
Index-Params
{"m":16,"ef_construction":128,"ef_search":64}
Dataset
S0 · 550.000 Vektoren · dim 1024 · Variante A · 2.00 GB
Queries
2.000 (Concurrency 1)

Lauf

Gestartet
11.06.2026, 10:02:46
Beendet
11.06.2026, 10:02:59
Dauer
7213s
Index-Bauzeit
364.28s
Index-Größe
4249.4 MB
CPU ⌀ / Peak
0.01 / 0.01 cores
RAM ⌀ / Peak
4278 / 4278 MB
Disk read (Paging)
5 MB
Disk write
2 MB
Pod-RAM-Limit
Default
Warmup-Queries
1.000
K8s
v1.31.4+k3s1
Nodes
4

Notizen

measured: in-cluster
decoupled: true
ingested_at: 2026-06-10T22:03:21Z
ingest_config: pgvector-S0-latency
insert_time_s: 217.97
n_vectors_actual: 550000
has_metadata: true
mem_limit_gb: null
pre_run_reset: skipped (decoupled, warmup-discard)
n_queries_executed: 1000
n_warmup: 1000
gt_file: ground_truth_filter_rating_gte_4_ids.npy
mode: measure-only