R: [Linux-cluster] Performance tuning help sought

[Kadlecsik Jozsef]

Hi,

On Tue, 4 Mar 2008, Bill Sellers wrote:

> 
> There are some TCP tuning parameters you can change in the /etc/sysctl.conf.
> Adding these made a noticeable improvement for us. Also, if you use any NFS,
> set your rsize and wsize = 32768 (NFS mount options) in a Linux 2.6 kernel.
> The noatime mount option appears to work well.  We have abandoned ext3 for
> filesystems > 1Tb in favor of reiserfs and jfs.   These other filesytems
> appear to give better performance in a cluster environment. If you use gfs as
> the filesystem, then the above point is moot.
> 
> http://dsd.lbl.gov/TCP-tuning/linux.html
> 
> http://www.nren.nasa.gov/tcp_tuning.html
> 
> http://nfs.sourceforge.net/nfs-howto/ar01s05.html

Thank you the tips, to all of you, NFS will be tuned for sure!

Digging into the mailing list archive brought out some GFS tuning options 
too. What is already done:

- statfs_slots set to 128
- statfs_fast set to 1	

What we are going to try is to set 

- glock_purge to 50
- demote_secs to 150

Best regards,
Jozsef
--
E-mail : kadlec at mail.kfki.hu, kadlec at blackhole.kfki.hu
PGP key: http://www.kfki.hu/~kadlec/pgp_public_key.txt
Address: KFKI Research Institute for Particle and Nuclear Physics
         H-1525 Budapest 114, POB. 49, Hungary