PostgreSQL pgbench performance regression in 2.6.23+
Hi guys!
Quoted 4 lines
Quoted 2 lines
I.e. i think it's a perfectly fine answer to say "if your workload needs batch scheduling, run it under
SCHED_BATCH".Yes, and this appears to be such a case.
Excuse me for interrupting you, but am I getting this right: to run effectively PostgreSQL with an active web app on the server a "system administrator Joe" has to know about batch scheduling?
Signature · 1 line
Anton Petrusevich
Quoted 3 lines · Anton Petrusevich
Excuse me for interrupting you, but am I getting this right: to run effectively PostgreSQL with an active web app on the server a "system administrator Joe" has to know about batch scheduling?
That's not true. The problem here is specifically with the pgbench program, which is a simple benchmarking tool included with PostgreSQL. pgbench already had known scalability problems due to its inefficient design and the new scheduler really doesn't play well with it by default. It's the pgbench program that needs the batch scheduling, not the database processes.
pgbench is certainly not representative of web application performance, and it wouldn't be the first time there was some OS tweaking required to get good results from such an artificial benchmark.
--
* Greg Smith [email protected] http://www.gregsmith.com Baltimore, MD