Hello Ben, Friday, September 29, 2006, 10:07:12 PM, you wrote:
BR> Hey Guys, BR> I'm trying to tune my NFS enviroment and have yet to make any BR> improvement, I was hoping someone could offer some experience in this situation. BR> Here's the problem: I've got a bunch of X4100 clients (NV_B43) BR> and a Thumper (NV_B43) NFS server using ZFS for storage. I'm BR> currently using NFSv3 TCP. Sequential read and write performance BR> is excellent (100MB/s+) but creating a large number of files, BR> namely untar'ing something, is just dog slow. Because people tend BR> to use tar more than they create 10G sequential files, people are BR> complaining about the horrible performance. Actually I belive this is NFS+ZFS problem and not NFS alone. You can easily check it - on nfs server set zil_disable to 1 (echo 'zil_disable/W 1'|mdb -kw' then export/import a pool. Now try again - you should see great performance increase in you case. If above works then try to upgrade to latest nevada as several fixes (bugs like syncing whole pool when only one file is synced, etc.) where introduced or just live with zil_disable set to 1 if you can. -- Best regards, Robert mailto:rmilkowski at task.gda.pl http://milek.blogspot.com
