Re: Pgbench performance tuning?

On Thu, Aug 4, 2016 at 6:53 PM, Andres Freund <andres(at)anarazel(dot)de> wrote:
> What's the config? Version? What activity does pidstat -d -l indicate?
> How much WAL was generated?

I know the specifics matter but I was also trying to avoid dumping too
much into the email.

The shared buffers is set to 16384 (128MB). Otherwise it's a default
config (For 9.4 and before I set checkpoint_segments to 32 as well).

Never seen pidstat before but pidstat -d looks like it prints very
similar output to the iostat output I was gathering already. There's
nothing else running on the machine.

I would have to rerun the benchmarks to measure the WAL output. Since
there's no replication and the database shut down cleanly it looks
like it recycled all the log files proactively and the last checkpoint
is in the earliest numbered xlog file.

I'll add checking the xlog position before and after pg_bench to the
script. I also wanted to use run the database under a binary calling
getrusage to report RUSAGE_CHILDREN for the database. It looks like
there's no stock program to do this but it shouldn't be hard to hack
one up.

--
greg