Subject: Re: multiple processors on NFS servers
To: Jonathan Stone <jonathan@dsg.stanford.edu>
From: Michael Richardson <mcr@sandelman.ottawa.on.ca>
List: tech-net
Date: 04/12/2000 21:19:54
>>>>> "Jonathan" == Jonathan Stone <jonathan@DSG.Stanford.EDU> writes:
    Jonathan> In message <200004130013.UAA06681@pzero.sandelman.ottawa.on.
    ca> Michael Richardson writes:
    Jonathan> [nfs server]

    >> Advice?

    Jonathan> what kind of network?

    Jonathan> These days, even cheap disks sustain 15 Mbytes/sec for
    Jonathan> large-ish reads. (I think high-end SCSI disks are about twice
    Jonathan> that).

    Jonathan> Even a 166MHz Pentium could handle the copying and checksumming
    Jonathan> to saturate a 100Mbit NIC. So unless your NFS servers have
    Jonathan> multiple NICs, or gigabit Ethernet, your bottleneck is likely
    Jonathan> to be 100mbit media, not CPU or disk.

  We currently have a pair of 450Mhz PIIIs and 256Mb ram, running:

    Jonathan> (I concur on avoiding Linux, btw.  Linux 2.2.x finally got
    Jonathan> in-kernel NFS.  It works, mostly, with Linux clients, but (as
    Jonathan> of 2.2.12) for Solaris or BSD clients, it's busted. And once
    Jonathan> you fall back to userspace NFS, the performance is as horrible
    Jonathan> as it ever was.)
  
  Well, it is busted for 2.2.12 with Linux clients as well. Big files
certainly do it. 2.2.14 is supposedly better, but we haven't seen much. 
We can kill the file server, as in knfsd processes die in the kernel by
compiling our source tree from a Solaris box. 

]      Out and about in Ottawa.    hmmm... beer.                |  firewalls  [
]   Michael Richardson, Sandelman Software Works, Ottawa, ON    |net architect[
] mcr@sandelman.ottawa.on.ca http://www.sandelman.ottawa.on.ca/ |device driver[
] panic("Just another NetBSD/notebook using, kernel hacking, security guy");  [