Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!cs.utexas.edu!rice!sun-spots-request
From: jaffa@tut.fi (Salo Jari)
Newsgroups: comp.sys.sun
Subject: Re: Sun 4/390 server
Keywords: Networks
Message-ID: <4987@brazos.Rice.edu>
Date: 14 Feb 90 10:01:32 GMT
Sender: root@rice.edu
Organization: Sun-Spots
Lines: 35
Approved: Sun-Spots@rice.edu
X-Refs:  Original: v9n38
X-Sun-Spots-Digest: Volume 9, Issue 33, message 13

On 8 Feb 90 15:33:02 GMT,
rich@ndcvb.cc.nd.edu (Rich Sudlow) said:

| We recently installed a Sun 4/390 server with 30 diskless clients
| containing 16 MB of memory. However we found that it was fairly slow and
| performance dropped considerably with just with just a few clients
| running. It appears that many times the nfsd's just start to run for no
| reason at all (Yes we did reboot the clients with the server). Has anyone
| had similar problems of this sort??

We have a bit similar environment to yours:

4 sun 3/260 servers (16 Mb each) and about 30 workstations
loading their disks through NFS. Sun OS 4.0.3.

We had the problem of nfsd:s getting in to a deadlock every now and then.
they seemed to be all in inode-wait-state.  With pstat -T it seemed as if
there really were no inodes free.  What we did was to triple the size of
inode-entries in system kernel of the servers. it is configurable in
config.c.

The original was:

int ninode = (NPROC + 16 + MAXUSERS) + 64;

which is a bit small as count of open files is determined using that very
same expression ... and as NFS (if I have understood it right) doesn't
really open files itself, but does use inodes a lot.

So far we have not (yet) had the situation occuring again.

Jari Salo                                           #  The  knack of  flying
Tampere Univ of Technology / Software Systems Lab   #  is  learning  how  to
PO box 527                  #  Mail: jaffa@tut.fi   #  throw yourself at the
SF-33101 Tampere, Finland   #  Tel : 358-31-162932  #  ground  and  miss.