[OpenAFS] Solaris 10, OpenAFS, and Firefox 3.0.3

Kris Kasner tkasner@gmail.com
Wed, 12 Nov 2008 11:35:26 -0800

Have you tried moving to memcache just for fun?

I have reason to believe that due to the change in the VFS interface
in Solaris 10, disk based cache has a memory leak and a huge
performance hit.

Solaris SPARC and x64 with disk cache both falter under the same load
as a linux system with the same OpenAFS version (same specs on the
Solaris x64 and the linux box). The SPARC system can handle more load
than either of the x64 systems when we're using memory cache.
These systems are a corner case - they are rsync servers for 1000+
systems that need copies of AFS data, but are not AFS clients. Each of
these servers have about 100-200 clients rsync'ing data every 15
minutes or so.

It might be worth it to try the memcache just to see if it makes a
difference in your environment.

Good luck..

On Wed, Nov 12, 2008 at 10:25 AM, Jeff Blaine <jblaine@kickflop.net> wrote:
> I'm throwing this out there because I'm getting close
> to the end of my ideas on this problem machine.  Just
> in case someone has experienced something similar or
> has any ideas.
> * OpenAFS 1.4.7
> * SunOS 5.10 Generic_127127-11 sun4u sparc SUNW,A70
> * Firefox 3.0.3 for Solaris 10 (Contributed by Sun)
> * Thunderbird ^^^^^^^^^^^^^
> PROBLEM : Excruciating slowness of both Firefox and
> Thunderbird and, eventually, the entire box.  It
> can take over 5 minutes for the apps to display
> on screen.
> Running Firefox and Thunderbird *outside* of AFS works great.
> Running the same Firefox and Thunderbird *from AFS* on other
> machines works fine (same OS rev, same AFS rev, etc).
> 1.  Box was completely re-imaged and examined closely for
>    AFS oddities a few days after the problem was found.
>    It did not help.
> 2.  'fs getserver' shows proper local fileservers as
>    preferred for the binaries.
>      fs1.foo.com             40004
>      fs2.foo.com             40014
>      fs3.foo.com             20009
>      fs4.foo.com             40010
>      fs5.foo.com             40006
>      fs6.foo.com             20010
>      server fs1.foo.com partition /vicepa RW Site
>      server fs6.foo.com partition /vicepa RO Site
>      server fs2.foo.com partition /vicepa RO Site
>      server fs5.foo.com partition /vicepa RO Site
>      server fs3.foo.com partition /vicepa RO Site
>      server fs1.foo.com partition /vicepa RO Site
> 3.  No network interface errors and all other networking
>    appears fine/fast.
> 4.  Local disk-based cache like all of our other hosts.  60MB.
> Stumped as of now.
> _______________________________________________
> OpenAFS-info mailing list
> OpenAFS-info@openafs.org
> https://lists.openafs.org/mailman/listinfo/openafs-info