Hi,
On 2016-04-06 21:58:50 -0400, Robert Haas wrote:
> I spent a lot of time testing things on power2 today
Thanks for that!
> It's fairly mysterious to me why there is so much jitter in the
> results on this machine. By doing prewarming in a consistent fashion,
> we make sure that every disk run puts the same disk blocks in the same
> buffers. Andres guessed that maybe the degree of associativity of the
> CPU caches comes into play here: depending on where the hot data is we
> either get the important cache lines in places where they can all be
> cached at once, or we get them in places where they can't all be
> cached at once. But if that's really what is going on here, it's
> shocking that it makes this much difference.
I'm not sure at all that that is indeed the reason. I think it'd be
quite worthwhile to
a) collect perf stat -ddd of a slow and a fast run, compare
b) collect a perf profile of a slow and fast run, run perf diff
c) add perf probes for lwlocks and spinlocks where we ended up sleeping, check whether the two runs have different
distributionof which locks ended up sleeping.
I'm also wondering if the idea from
http://archives.postgresql.org/message-id/20160407082711.q7iq3ykffqxcszkv%40alap3.anarazel.de
both with/without preloading, changes variability.
Greetings,
Andres Freund