Re: spinlocks on HP-UX

Robert Haas <robertmhaas(at)gmail(dot)com> writes:
> I am a bit surprised by your test results, because I also tried x86_64
> with an unlocked test, also on pgbench -S, and I am pretty sure I got
> a regression. Maybe I'll try rerunning that. It seems possible that
> the x86_64 results depend on the particular sub-architecture and/or
> whether HT is in use, which would be kind of a nuisance.

Well, if you consider Opteron as a sub-architecture of x86_64, that was
already true the last time we did this. So far there have not been
cases where something really good for one implementation was really bad
for another, but someday we'll probably hit that.

> Also, did you happen to measure the amount of user time vs. system
> time that your test runs used?

Did not think about that. I was considering how to measure the average
context swap rate over each run, so that we could keep an eye out for
the "context swap storm" behavior that's the usual visible-in-top
symptom of these sorts of problems. But it'd have to be automated;
I'm not going to keep my eyes glued to "top" output for several hours.

I'd be happy to re-run these tests with any RHEL-compatible measurement
scaffolding somebody else provides, but if I have to write it, it
probably won't happen very soon.

> If this is on Linux, I am surprised
> that you didn't get killed by the lseek() contention problem on a
> machine with that many cores.

Hm ... now that you mention it, all of these tests have been using
the latest-and-greatest unreleased RHEL kernels. Maybe Red Hat already
fixed that contention problem in their kernel? Have you got a RH
bugzilla number for the issue?

regards, tom lane