Yaniv,
I'm adding dtrace-discuss to this email for reasons that will be obvious
immediately :-) - see below
Yaniv Aknin wrote:
> When volumes approach 90% usage, and under medium/light load (zpool
> iostat reports 50mb/s and 750iops reads), some creat64 system calls take
> over 50 seconds to complete (observed with 'truss -D touch'). When doing
> manual tests, I've seen similar times on unlink() calls (truss -D rm).
>
> I'd like to stress this happens on /some/ of the calls, maybe every
> 100th manual call (I scripted the test), which (along with normal system
> operations) would probably be every 10,000th or 100,000th call.
I'd suggest you do something like this (not tested, so syntax errors etc
may be lurking; I'd also suggest you get the DTrace guide off of
opensolaris.org and read the chapter about speculations):
#!/usr/sbin/drace -Fs
int limit ONE_SECOND /* you need to replace this with 10^9, I think)
syscall::creat64:entry
{
self->spec = speculation();
speculate(self->spec);
self->ts=timestamp();
self->duration = 0;
}
fbt:::entry,
fbt:::return
/self->spec/
{
speculate(self->spec);
}
syscall::creat64:return
/self->spec/
{
speculate(self->spec);
self->duration = timestamp() - self->ts;
}
syscall::creat64:return
/self->duration > limit/
{
commit(self->spec);
self->spec = 0;
}
syscall::creat64:return
/self->spec/
{
discard(self->spec);
self->spec = 0;
}
you may need to use a different timestamp (walltimestamp?); and perhaps
you'll want to somehow reduce the number of fbt probes, but that's up to
you. I hope you can take it from here.
cheers
Michael
--
Michael Schuster Sun Microsystems, Inc.
recursion, n: see 'recursion'
_______________________________________________
zfs-discuss mailing list
zfs-discuss@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
