Re: knfsd fixed, but SLOW

Olaf Kirch (okir@monad.swb.de)
Sat, 12 Apr 1997 11:27:19 +0200

On Fri, 11 Apr 1997 23:29:10 EDT, "Steven N. Hirsch" wrote:
> To my chagrin, the throughput with the new server is horrendous. Haven't
> had much time to look into it, but there seems to be almost constant disk
> activity - almost as if nothing was being cached and every byte being
> flushed as soon as it was received by the server.

This is because by default, nfsd writes all data it receives from the
client synchronously. The NFS spec mandates that the server commits all
data to stable storage before returning a success result to the client.
This will hit you particularly hard if you write data from the old NFS
client which does not buffer writes.

You can either turn off sync writes completely by exporting your fs's with
the async option (which is unethical according to Sun Micro gospel)
or try gathered writes (legel, give a slight improvement, but only with
clients that do write buffering).

kexportfs foo:/usr -o async,rw,no_root_squash # buffer all writes
kexportfs foo:/usr -o wdelay,rw,no_root_squash # gathered writes

Olaf

-- 
Olaf Kirch         |  --- o --- Nous sommes du soleil we love when we play
okir@monad.swb.de  |    / | \   sol.dhoop.naytheet.ah kin.ir.samse.qurax
             For my PGP public key, finger okir@brewhq.swb.de.