> We use one Solaris NFS server to serve several mount > points. One of the mount points serves a client that > writes a large number of different small filles for > long periods. This seems to hurt the performance of > the clients of the other mount points. > > The server has no lack of cycles, network, or i/o > bandwidth. The issue is not disk contention. > > I imagine the issue is that the busy mount point is > thrashing the buffer cache, and the authcache. One > effect we definitely see is a substantial netgroup > lookup load on the ldap server, presumably because of > the authcache. > > Is there any way to limit the effects that the busy > mount point has on the others. An SRM kind of > approach?
Take a look at nfsstat and the /etc/default/nfs file, and try up the value for Set connection queue length for lockd over a connection-oriented transport for the NFSD_LISTEN_BACKLOG=32 change to 64. Then next where your issue most likely maximum number of concurrent NFS requests, this is equivalent to last numeric argument on nfsd command line by adjusting NFSD_SERVERS=16 to 64 , then you can set connection queue length for lockd over a connection-oriented transport to LOCKD_LISTEN_BACKLOG=32 to 64, the memory resources are very low on system with more the 512meg of ram, then years ago when a nfs server only had 64meg or 96meg the values needed to more closely watched and limited. stop and restart the svcs nfs services svc:/network/nfs/server. This message posted from opensolaris.org _______________________________________________ opensolaris-discuss mailing list [email protected]
