The disk layout is RAIDZ2. The issue isn't present on ZFS local. The problem is an NFS tuning issue. It is possible that there is a ZFS-NFS issue, but I find that somewhat unlikely although I can't rule it out.
When I watch NFSD thread count during access I notice that only 5 threads open during a 12MB untar via NFS. That points to a couple of possible things that could be better, such as better caching on the client side to allow more concurrent threads running back to the NFS server allowing the actual IO to be less synchronous, but I'm not really an NFS expert. I might be time for me to consider CacheFS on the client side, but I think there is a much simpler tuning solution that I'm simply unaware of or haven't attempted yet. benr. This message posted from opensolaris.org _______________________________________________ perf-discuss mailing list perf-discuss@opensolaris.org
