On Tue, 27 Jan 2004, John Reuning wrote: > Looks like this thread didn't receive much attention. Here are a few > more suggestions: > > 1. Check out the netapp nfs linux client performance tuning document. > http://www.netapp.com/tech_library/3183.html > > Reviewing it, I had read an earlier version.
> 2. There's a max number (256, I think) of outstanding r/w requests per > mountpoint in 2.4 kernels. If you think this might be an issue, try > breaking up the nfs exports. So are you saying if I have the following mount points 10.1.1.2:/vol0/nfs1 /export/home 10.1.1.2:/vol0/nfs2 /export/home2 That on each linux NFS client there could only by one r/w request on to each mount point. I'm not sure I complete understand the ramifications of this point; but would like to because we have a large number of directories under each mount point so this may be important. > > 3. Check out research papers at the citi site for some excellent > technical & research information. > http://www.citi.umich.edu/projects/nfs-perf > Will investigate, was not aware of this site. Thanks. > 4. If you're using redhat kernels, try building the latest in the RHEL > 2.1 2.4.9 series or 2.4.18 series. Depending on the application and the > system load, older kernels sometimes hold up to stress better. > We are using Redhat 7.1 but with latest kernels 2.4.24; are are now also using NFS read-ahead patch which seems to help with load times. But still having some slow updates with some shopping cart issues. > 5. I'm sure the netapp folks will have addressed this, but make sure > your network links are fast and clean. > We haven't found any network issues to date; it's a private network that the Netapp filer is on so I don't believe it is on the network. But I'm not ruling anything out. I will continue to monitor stats. > And for more basic linux nfs information, there's always the nfs howto. > http://nfs.sourceforge.net/nfs-howto/ > I've visisted it a few times; and will continue to review things. I'm on the mailing list. > Finally, one question: is the shopping cart a write-heavy application? > Netapps are raid-3-ish storage, right? Have the netapp folks given > recommendations on raid volume size? It seems, one of the shopping cards seems to be both write and read intensive. We are still trying to determine if this can be an issue with the application itself. We are working with netapp to see if we need to possible change our raid volume sizes based on some more current stats of our disk utilization, since we have found we can put quite a load on the box :-) But the box is handling I'm happy to say. Thanks for the feedback. I'm sure it will help. -- TriLUG mailing list : http://www.trilug.org/mailman/listinfo/trilug TriLUG Organizational FAQ : http://trilug.org/faq/ TriLUG Member Services FAQ : http://members.trilug.org/services_faq/ TriLUG PGP Keyring : http://trilug.org/~chrish/trilug.asc
