On 2015-07-05 11:05:28 -0400, Robert Haas wrote:
> More broadly, I don't really know how to test this patch and show when
> it helps and when it hurts. And I think we need that, rather than
> just a theoretical analysis, to tune the behavior. Heikki, can you
> describe what you think a good test setup would be? Like, what
> workload should we run, and what measurements should we gather to see
> what the patch is doing that is good or bad?
I think a good start would be to graph the writeout rate over several
checkpoints. It'd be cool if there were a better way, but it's probably
easiest to just graph the number of bytes written (using iostat) and the
number of dirty bytes in the kernel. That'll unfortunately include WAL,
but I can't immediately see how to avoid that.