PostgreSQL 18 lower TPM than PostgreSQL 17.7 in HammerDB TPC-C

[vijivignesh15112013-crypto]

Hi Team,

We are running HammerDB TPC-C benchmarks on PostgreSQL 17.7 and PostgreSQL 18, and we consistently see lower TPM with PostgreSQL 18 under the same test conditions.

Results summary:

PostgreSQL 17.7 → ~2.3M TPM

PostgreSQL 18 → ~2.0M TPM

The results are repeatable across multiple runs

Hardware

48 CPU cores
128 GB RAM

HammerDB

Version: 5.x
Benchmark: TPC-C
Driver: timed
Warehouses: 500
Virtual Users: 96
Ramp-up: 5 min
Duration: 3 hr
Keying/Thinking time: disabled

We used the same PostgreSQL configuration for both versions, except for the io_method.
All available io_method options were tested, and the behavior remained the same.

Questions

Are there any known performance changes or regressions in PostgreSQL 18 that could impact OLTP/TPC-C workloads?
Are there new tuning recommendations specific to PostgreSQL 18?
Has anyone else observed similar behavior with HammerDB on PostgreSQL 18?
Does HammerDB require any PostgreSQL 18–specific driver changes?

Thanks in advance.

[sm-shaw]

Hi, great finding and this is exactly what HammerDB is for! This is the very definition of a benchmark a repeatable workload so when you change a variable and change in performance can be attributed to the change you made, in this case the PostgreSQL version.

So your questions are best directed to a PostgreSQL forum rather than here and your next step is to look deeper into PostgreSQL performance to identify the cause of the difference. The HammerDB PostgreSQL metrics can be very useful here: https://www.hammerdb.com/blog/uncategorized/how-to-get-hammerdb-postgresql-metrics-up-and-running-fast-with-pg_sentinel/
In particular look at the event window to see if an event has increased between versions.

[vijivignesh15112013-crypto]

Thank you for the reply . will check