On Mon, 9 Jul 2007, Carl Horne wrote:
> Hi,
>
> We have a number of Sun SunFire V40z amd64 servers that were all bought
> around the some time and have the same hardware. I have been involved
> with a project that requires OpenBSD and I have been re-OSing then with
> OpenBSD 4.1 (amd64 mp). I have a mystery that I have not been able to
> figure out. The problem is the disk I/O is really slow and it pegs both
> processors when doing heavy I/O. I spent weeks trying to figure out why
> it was doing this but because of dead lines and because the end product
> was a fw/router I ended up rolling one out to production and had used
> another server running Solaris to do heavy disk I/O stuff. Now a few
> months later I built a new one and it does not have the disk I/O issue
> even though I installed it with the same boot cd and the same local copy
> of the distro.
>
> Since the hardware was the same I put the newly OSed drive in one of the
> other servers that has the problem and to my surprise it still worked
> great with no I/O issues. I have built a number of these systems and
> only this magical one works right and doesn't have the I/O issues no
> matter what server I put the drive in. I have spent some time trying to
> see if I could find any difference between the two installs but
> everything looks the same (file sizes, dmesg, etc). What could be
> causing the issue? How can I constantly build servers without using dd
> that doesn't have the problem?
It could be a disk write cache issue.
-Otto
>
> Test:
> For testing I used a tar ball I had laying around. It is 135M and has
> 9044 files in it.
>
> Server that has the I/O issue:
> bash-3.1# time tar -xzf mod_ssl-new.tar.gz
> real 3m22.241s
> user 0m1.440s
> sys 0m1.750s
>
> Server that doesn't:
> bash-3.2# time tar -xzf mod_ssl-new.tar.gz
> real 0m12.380s
> user 0m0.330s
> sys 0m0.810s
>
> Thanks in advance,
> Carl
