Re: 2.5 profiling

On Thu, Nov 22, 2001, Henrik Nordstrom wrote:
> On Thursday 22 November 2001 02.13, Adrian Chadd wrote:
>
> > So far, running aufs, squid-2.5 has been able to hold up an 80req/sec
> > load (on a celeron 500). Any more than 80req/sec and I flatline the CPU
> > and cause squid to go into spiralling sudden death, but that may
> > also be due to the restricted CPU in my (current) polygraph testing
> > machine. I'll try it from a faster machine today.
>
> How far do you get if you disable caching?
>
> That CPU should be capable of a lot more than 80.

I'm sure it'll go past 80 with the CPU pegged - just the service times
will blow out a little. I'll run it on a more powerful polygraph
machine tonight when I get home.

> What is your drive configuration?

3x 9gig 7200? RPM ultra160 scsi disks. Configured using reiserfs,
and I'm using 6gb of each per storedir.

> > I currently can't debug the squid-aufs code under irix, as gdb doesn't
> > do threaded applications at all. I'll do a trial run in a couple of days.
>
> Under Linux before GDB knew about Linux Threads one could at least debug the
> main thread.. or attach the debugger specifically to the pid of some other
> thread. But I guess this is a positive byproduct of the way Linux Threads are
> implemented (1-1 kernel threads using what looks like processes to everyone

Yup. But the threads under IRIX are implemented .. well, they look like
they're implemented much, much differently. :)

I'll give it some thought and play some more over the next few days.

adrian
Received on Wed Nov 21 2001 - 21:21:37 MST