Jay Estabrook wrote:
> The "grabbed" message simply indicates when a process that was "stuck"
> actually got the lock. The "ticks" indicated are jiffies, of course,
> and on ALPHA there are usually 1024/second, so when a "grabbed"
> message is printed saying it waited for 6732 ticks, that's 6+ seconds.
> That seemed awful long to me when I first saw it, so I asked either
> Linus or Alan about it, and they said it was "normal" or "reasonable",
> so I stopped worrying. Perhaps it's time to ask again, though I'd be
> quite worried if I saw that much time on a 2.4.xx kernel... ;-}
>
Truth be told, that's a _really_ long time. We start to freak out when we see
locks in Tru64 have a waitsum of >2 seconds _total_ in an hour benchmark run. In
Tru64, that's considered easy pickin's for perf guys like me, where it's a simple
sign of bad lock design. But then again, Tru64 has probably in the area of an
order of magnitude (at least) more locks than Linux does, so one lock hogging
that much time is a much clearer sign than the more coarse granularity of Linux.
So what am I saying? Yes, it is a bad thing from a perf perspective, and only
gets worse as the systems get bigger and faster. That's why I try to keep on top
of the lockmeter code for Alpha (can be gotten from SGI, works now with a small
patch from me until John integrates it) - so we can make sure we always know what
is going on whenever possible. (Not helpful in this case, as it's at boot time,
but other than that...)
BTW - Greg, could you send me a couple of lines of the "spinlock stuck" and
"spinlock grabbed" messages? I'm curious which locks are getting in the way, and
if they're the same ones I see now and again...
- Pete
-- To unsubscribe: send e-mail to axp-list-request@redhat.com with 'unsubscribe' as the subject. Do not send it to axp-list@redhat.com
This archive was generated by hypermail version 2a22 on Sat Jul 1 05:31:30 2000 PDT
Send any problems or questions about this archive to webmaster@alphalinux.org.