Page MenuHomeSoftware Heritage
Paste P489

warm cache cassandra vs postgresql read benchmark
ActivePublic

Authored by vlorentz on Aug 8 2019, 12:16 PM.
Benchmark results for content_find:
hash_algo = sha1 (sample size=248):
avg cassandra = 5 ms
avg postgres = 14 ms
hash_algo = sha1_git (sample size=242):
avg cassandra = 6 ms
avg postgres = 12 ms
hash_algo = sha256 (sample size=224):
avg cassandra = 6 ms
avg postgres = 14 ms
hash_algo = blake2s256 (sample size=286):
avg cassandra = 6 ms
avg postgres = 13 ms
Benchmark results for revision_get (sample size=1000):
avg cassandra = 3 ms
avg postgres = 19 ms
Benchmark results for directory_ls (sample size=1000):
avg cassandra = 52 ms
avg postgres = 51 ms
Benchmark results for release_get (sample size=1000):
avg cassandra = 2 ms
avg postgres = 18 ms
Benchmark results for snapshot_get (sample size=1000):
avg cassandra = 4 ms
avg postgres = 27 ms

Event Timeline

vlorentz changed the title of this paste from cassandra vs postgresql read benchmark to warm cache cassandra vs postgresql read benchmark.EditedAug 8 2019, 12:19 PM

Rough test benchmark of read access from the cassandra test cluster (4 desktops) and the postgresql replica DB on Somerset.

Keep in mind that:

  • Hardware is very different
  • Cassandra tables have about 5 times less objects (10 times for directories)
  • I ran this on a computer with 2.3ms ping to postgresql, 0.3ms to cassandra
  • Cache is warm on both cassandra and postgresql

Script: P490