On Thu, Jul 07, 2011 at 08:34:20PM -0700, Robert Gordon wrote: > stopping at starting nfsd will pick up the new value. you never > mentioned what version of solaris you are seeing this on .. also note > that 16 is the default (but i thought it was changed, or i opened a > bug to change it .. (i worked in the NFS Group for a few years)) > > look for reties in nfsstat output if you suspect things are getting > dropped, based on your 2200 ops a second i doubt it but then i don't > know what the server machine is ;) > > NFSD_LISTEN_BACKLOG is the queue depth of pending accepts on the > listen() socket for NFS, i suspect you have 30 client boxes, the > client will keep connections open, so it's not generally a problem > for small work loads like you appear to have .. > > hth, > Robert.
Whoops -- Solaris 10 U9 is what we're running. :) This box was set up quite a while ago, so 16 was likely the default at that time. I assume we'd need to look for those retries on the client side. Unfortunately, it doesn't seem like the ESXi hypervisor presents or makes these stats available any longer (we have to query via a vMA appliance now instead of via a local shell). Based on further investigation from suggestions in this thread, I would also agree that the backlog isn't the issue here. We have an even less busy server though with max threads set to 1024 and it has ~50 active threads at any given time, so I'm guessing _this_ box would definitely benefit some from moving beyond the default of 16. :) Thanks! Ray _______________________________________________ nfs-discuss mailing list [email protected]
