Virtual Memory tuning and Buffers

["Luke Miller" <millerlu@integraonline.com>]

We have a system that is running SUSE 8 kernel 2.4.21-273.  Recently we have
seen some interesting behavior with the virtual memory system.  When buffer
memory gets low (below 10 MB) the system re-allocates memory causes buffer
memory to go up to around 500 MB.  The re-allocation can take up to 30
minutes.  This system is an NFS server and during the re-allocation all the
NFSd's go to IO wait and the NFS clients gets delays trying to access files
on the NFS volumes.  I have generated some graphs that show this pretty
clearly, they are here:

http://www.integraonline.com/~millerlu/memory/

What I am trying to figure out is what is causing this memory allocations
and what can be done to tune them so it doesn't impact our production
applications when it happens.

Due to a clustering and file system product, Polyserve, we are unable to
upgrade to the 2.6 kernel at this time.

Any help would be appreciated.

Thanks,

Luke


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/