On Mon, 7 Nov 2005, Daniel Widyono wrote:

>> symptoms.  Some have suggested either increasing number of nfsd's to 300
>> or greater, or abandoning TCP altogether and reverting to UDP, which
>
> 4-way... I've heard (sorry don't remember source but it was found via google)
> you should limit yourself on linux to max 8 nfsd's per processor, that would
> mean 32 processes for 600+ mounts.
We are hyper-threaded with 8 effective processor, does that make a 
difference?  (We found the same thing in Google).


Yikes.  We have about 15 mounts per
> server process, 8 processes per processor, also gives poor yield.  We're
> migrating to a SAN with GFS to "solve" this issue (among several other
> issues).

We are experimenting with doing the same.
>
> Is your server's network card stable and not saturated?
>
We are having some packet receive errors but nobody can tell us why.
If there are other ideas on how to tell whether the network card
is saturated, let me know.


>> 3) will change mount options, remove bg, add noac,rsize=32768, wsize=32768
>
> 32768?  Do you have the networking equipment to handle that without
> fragmenting?  Are you already using jumbo frames with your 8192 packet size?
>

No, no jumbo frames.. NFS howto says with TCP this size is OK.

Thanks

Steve


> Good luck,
> Dan W.
>

-- 
------------------------------------------------------------------
Steven C. Timm, Ph.D  (630) 840-8525  [log in to unmask]  http://home.fnal.gov/~timm/
Fermilab Computing Div/Core Support Services Dept./Scientific Computing Section
Assistant Group Leader, Farms and Clustered Systems Group
Lead of Computing Farms Team