Slide 13
Slide 13 text
How to measure query performance
Benchmark
● Instance: m5ad.4xlarge (16 vcores, 64GB RAM, local SSD)
● Standard pgbench with 24 clients, 12 jobs, scale=5000 (+100GB) and --log
● 1 warm up run, 3 runs of 10 minutes:
○ Tps: averaged over three runs
○ Latencies: computed percentiles over all results
● Cache cleaning, vacuum, checkpoints before every run. No autovacuum.
● Config:
https://postgresqlco.nf/manage/config/public/bd513ff0-e16b-4650-9918-a1452437a030
● Source code: https://gitlab.com/-/snippets/2290113