Umbrella bug for HammerDB tproc-c perf problems

I will use this as an umbrella bug to cover my performance reports using tproc-c from HammerDB.

First up is results from a small server (8 CPU cores, AMD SMT disabled, 32G RAM, 1 SSD) and the spreadsheets are here. The spreadsheet has results for MariaDB (mostly LTS) releases from 10.2 through 12.3, MySQL 5.6, 5.7, 8.0, 8.4, 9.4, 9.5 and Postgres 12 through 18.

A blog post is here but that is limited to MySQL and Postgres.

Notes:

• vu means "virtual users", number of concurrent users for HammerDB
• wh or w means "warehouses" for HammerDB

The summary is:

• for old versions (Postgres 12, MySQL 5.7, MariaDB 10.2) - MySQL and MariaDB were a bit faster than Postgres at some concurrency (vu=6) with workloads that do more IO (w=1000 and 2000). Otherwise Postgres was a lot faster. Results are in the pn53.old tab
• for new versions (Postgres 18, MySQL 8.4, MariaDB 12.2) - Postgres is always faster than MySQL and MariaDB. MySQL has larger regressions when more CPU-bound (vu=1 and w=100) while MariaDB has larger regressions at vu=6 and w=1000 or w=2000. Results are in the pn53.new tab. It is easier to see the differences in MySQL vs MariaDB in the pn53.new.my.ma tab.

To see changes over time for MariaDB 10.2 through 12.3 see the pn53.ma tab.

• the largest regressions (about 35%) arrive in 11.4.9 for vu=6 with wh=1000 or =2000 and those remain in 12.3
• otherwise, throughput is between 10% and 15% less in 12.3 vs 10.2 and in those cases the regressions are more incremental (per-release) rather than one huge drop

To see changes over time for MySQL 5.6 to 8.4 see the pn53.my tab.

• the regressions here are large from 5.6 to 5.7, and then from 5.7 to 8.0. The regressions after 8.0 are much smaller
• regressions are largest for the most CPU-bound workload (vu=1, wh=100) and there is no regression for the most-concurrent/most-IO-bound workload (vu=6, wh=2000).

To see changes over time for Postgres 12 to 18 see the pn53.pg tab.

• Postgres 18 is faster than Postgres 12
• Postgres avoids regressions while MariaDB and MySQL have many