On Tue, May 11, 2010 at 09:55:18AM -0400, Steve Shockley wrote:
> On 5/9/2010 11:28 PM, Claus Assmann wrote:
> >PS: you might want to run some of those disk I/O benchmarks
> >to determine the number of IOPs your system can provide.
>
> Thanks, everyone, for your help. I followed Nick's advice and went
> in the server room to watch the lights, and they're really not
> blinken that much. I had expected them to be on constantly.
>
> I tried running bonnie -s 1024 while watching the output of iostat.
> The iostat -w 5 -d output at "rest" (relaying mail, LA 0.9-1.5) is
> 8-16 KB/t, 0-15 t/s, 0.01-0.10 MB/s. During the run it seemed to
> max out at ~10 MB/s on writes and ~32 MB/s on reads.
>
> I also ran Jeff Ross' first dd test:
>
> > jr...@varley:/var/postgresql $ sudo time dd if=/dev/zero of=big_file
> > bs=8k count=1024000
> > 1024000+0 records in
> > 1024000+0 records out
> > 8388608000 bytes transferred in 276.573 secs (30330468 bytes/sec)
> > 276.60 real 0.17 user 60.39 sys
>
> # time dd if=/dev/zero of=big_file bs=8k count=1024000
> 1024000+0 records in
> 1024000+0 records out
> 8388608000 bytes transferred in 204.444 secs (41031301 bytes/sec)
> 3m25.64s real 0m0.57s user 0m49.42s system
>
> So, I got slightly better performance out of these 10k disks.
> iostat said I was doing about 20mb/sec. I'll have to run the
> /dev/null test later so my users don't yell at me about delayed
> emails.
>
> I have no milters running. I do have TLS enabled, but I wouldn't
> think that'd have an impact on a 3 GHz machine, and if it did I'd
> expect high CPU use. I'll keep looking, but at this point I'm
> thinking I'm not disk bound.
>
> For completeness, bioctl and dmesg pasted below. I'm running
> 4.7-stable with patch 004. The dmesg says it's a Smart Array 64xx,
> but I'm pretty sure it's a 6i, if that matters.
Look at top, do you have particularly high cpu usage due to interrupts?
-0-
--
In seeking the unattainable, simplicity only gets in the way.
-- Epigrams in Programming, ACM SIGPLAN Sept. 1982
