Benchmarks

Summary

On average, pg_uuidv7 is nearly as fast as the native gen_random_uuid() function in Postgres. It is perhaps 2% slower in a worst case scenario, however run-to-run variations of pgbench were >2%, which likely means that in a real system the performance impact of pg_uuidv7 is negligible.

Methods

Performance benchmarks were evaluated using pgbench. The following functions were benchmarked:

  • the native gen_random_uuid() function which is built in since Postgres 13
  • the uuid_generate_v7() function from this extension (pg_uuidv7)
  • a pure sql version of uuid_generate_v7() from kjmph

Results

``` pgbench --client=8 --jobs=8 --transactions=200000 --file=${TEST}.sql

-- SELECT gen_random_uuid(); scaling factor: 1 query mode: simple number of clients: 8 number of threads: 8 maximum number of tries: 1 number of transactions per client: 200000 number of transactions actually processed: 1600000/1600000 number of failed transactions: 0 (0.000%) latency average = 0.076 ms initial connection time = 5.039 ms tps = 105898.857590 (without initial connection time)

-- pg_uuidv7 C extension scaling factor: 1 query mode: simple number of clients: 8 number of threads: 8 maximum number of tries: 1 number of transactions per client: 200000 number of transactions actually processed: 1600000/1600000 number of failed transactions: 0 (0.000%) latency average = 0.076 ms initial connection time = 4.553 ms tps = 104904.866366 (without initial connection time)

-- sql function r24 scaling factor: 1 query mode: simple number of clients: 8 number of threads: 8 maximum number of tries: 1 number of transactions per client: 200000 number of transactions actually processed: 1600000/1600000 number of failed transactions: 0 (0.000%) latency average = 0.089 ms initial connection time = 4.965 ms tps = 89461.762254 (without initial connection time) ```