On 03/04/2011 10:24 PM, Brian Hirt wrote: > Hi, > > I have recently set up a two node cluster (on ubuntu 10.04LTS servers) using > drbd, nfs & pacemaker/heartbeat. Everything is working well so far. I > have some questions about best practices for the clients that are mounting > the volume via NFS and hope people here have some advice. Specifically, I'm > curious about some of the NFS mount options like proto=udp vs proto=tcp, > hard vs. soft, nfsvers=3 vs. nfsvers=4, appropriate timeo= values, should > intr be set, etc. I've already noticed that during failover with a low > timeout and soft mount, i can get NFS errors while the shared ip switches > over, so I'm pretty sure I want to use hard. Others options I'm not so sure > about. I've read some conflicting statements from various people on > different blogs/mailing lists.
Hi, I cannot comment on v3 vs. v4 as I haven't tapped into the latter yet. UDP should be more robust in failover situations, but I believe TCP yields better performance. You may want to be prepared to have all your clients perform a remount after you fail your server. As far as I understand, not using "intr" is saying "I want processes to be blocked in D state when there are NFS problems" - which I usually don't want to say. async is the most important setting performance-wise as far as I'm concerned. I never played with timeout settings. You may want to raise readsize and writesize above their defaults as well. I believe that in most situations, 1MB or at least 32k is more prudent than the default (8k?). HTH, Felix _______________________________________________ drbd-user mailing list [email protected] http://lists.linbit.com/mailman/listinfo/drbd-user
