CODEWITHSUNDEEP
performancepostgresql
A.A
A.A

Reputation: 4101

PostgreSQL Scalability: Towards Millions TPS

I read this article http://akorotkov.github.io/blog/2016/05/09/scalability-towards-millions-tps/

I try this benchmark in stronger machine but I achieve 20K transactions per second in max , how is possible to achieve 400K - 1.8M transactions per second? it is fake?

Benchmark graph

Upvotes: 3

Views: 4489

Answers (1)

Alexander Korotkov
Alexander Korotkov

Reputation: 81

I'm not sure if it's worth commenting 4 years after :) That's not fake. The usage of the term TPS for read-only pgbench transactions might be confusing. It would probably better to use the term QPS. However, the post explicitly says it's read-only pgbench benchmark and even the graph title says it's "pgbench -S". Nowadays, it's quite easy to reproduce. For instance, you can get c5d.18xlarge AWS machine and get even higher numbers on modern PostgreSQL. Nevertheless, I believe it's possible that you get just 40k read-only TPS on some "stronger machine". For instance, the "stronger machine" might appear to be NUMA with a very slow interconnect. Such a machine would be bad for PostgreSQL, it would be stuck there. The first steps for understanding what's going on would be collecting perf profile and statistics of PostgreSQL wait events while running the pgbench.

Upvotes: 3

Related Questions