>>> On Wed, 19 Oct 2005 16:52:03 -0700, Dave Pifke
>>> <[EMAIL PROTECTED]> said:

dave> I'm doing some performance tweaking on a web server that
dave> has a 2TB JFS partition,

Ha! that probably means some RAID arrangement. I would be happy
to know how long a full 'jfs_fsck -f' takes on that filesystem,
if you have ever had to do it.

dave> and I'm curious about a memory usage pattern I'm seeing
dave> with regards to the Buffer/Cache usage reported in
dave> /proc/meminfo.

dave> The system has 2GB of RAM.  At the moment, ~117MB is free,
dave> ~1.5GB is being used for buffers(!), and ~180MB is cached.

As another reply says the specific issue is probably just a
little bug causing a leak, but there are other amusing details,
so lets go on.

dave> I'm concerned that the relatively small amount of disk
dave> cache is hurting performance.

To me this seems a very strange concern. You have a cache that
is at best 0.1% (1/1000th) the size of the volume of data it is
supposed to cache and you are «concerned» about the difference
between 1/1,000th and 1/10,000th? Uhm, a bit strange.

File system caching except for the top levels of the tree and
some metadata is usually pointless anyhow. Some musings on this
topic here:

  http://WWW.sabi.co.UK/Notes/anno05-4th.html#051008
  http://WWW.sabi.co.UK/Notes/anno05-4th.html#051009

Even more so as the access pattern seems essentially random
among many small files:

dave> The machine uses Apache (mpm_worker) to serve up image
dave> files - a third thumbnails (<1kB), a third resized to the
dave> 50-100kB range, and a third originals (max 1MB). [ ... ]

dave> Directories are hashed year/month/day/hour with maybe 4000
dave> files in each.

I think that JFS has got fairly efficient tree directories,
perhaps having multiple directories is not essential.

Conversely one could explore storing the thumbnails in a
GDBM/Berkeley DB/... hash database and probably the 50-100KB
range images too.

Some example from a similar problem:

  http://lists.GLLUG.org.UK/pipermail/gllug/2005-October/055488.html

dave> The filesystem is mounted with -o noatime and I'm using
dave> the cfq elevator.

Good ideas, but also I would try the 'deadline' elevator and the
'noop' one too. They may be more appropriate (especially 'noop').

dave> The machine is fairly responsive, but I'm seeing a load
dave> average of between 2 and 3 with about 250 concurrent
dave> Apache requests

The load average may not be very relevant to what looks like an
IO/network bound server.

dave> - I'm obviously hoping for better performance. Ideas?
dave> Or is this about what one would expect given the system
dave> configuration and workload?

What is «this»? There is no indication of what performance is
like beyond «fairly responsive». It would be more useful to have
pretty vital details like average requests/s, MiB/s sent, files/s
accessed, and even just 10-20 lines of 'vmstat 1' while the
system is under load.



-------------------------------------------------------
This SF.Net email is sponsored by:
Power Architecture Resource Center: Free content, downloads, discussions,
and more. http://solutions.newsforge.com/ibmarch.tmpl
_______________________________________________
Jfs-discussion mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/jfs-discussion

Reply via email to