Re: Perf info & tools for NFS? (summary)

From: Bruce Hamilton (
Date: Mon Dec 18 1989 - 19:04:16 CST

Tom Herbst <Herbst.ESAE@Xerox.COM>: Barry Shein <>,
Software Tool & Die, 617-739-0202, 1330 Beacon Street, Brookline, MA 02146
presented a paper at Usenix last summer that defined the metric

Carl Rigney <cdr@amdcad.AMD.COM>: you can get nhfsstone [a program to
measure performance in NSFstones] from the Legato archive server by sending
mail with the subject "send unsupported nhfsstone" to;
as usual, you can mail "help" or "index" as well.

John Litvin <>: "If you would like to receive
regular information and bug fixes please send your name, and both your
Email and U.S. mail addresses to:

        Legato Systems, Inc.
        260 Sheridan Avenue
        Palo Alto, California 94306 or uunet!!nhfsstone-request

and we will add your name to the nhfsstone mailing list."

John B. Corwin <JCorwin.osbuNorth@Xerox.COM> has evaluated the Prestoserve
board from Legato Systems. This is an NFS accelerator board. It only
helps with WRITE operations. Corwin's servers only have 0-3% writes, so
the board didn't help.

Sunil Khatri <slcpi!slhcmg!slk@uunet.UU.NET>: "I'm currently using
something called netck... from Legato systems in Ca... ( ....i have [a
Prestoserver board]). netck... output looks like this"

dprg-1[34] netck -a dprg-1
dprg-1: high total packet rate
dprg-1: high input packet rate
dprg-1: high output packet rate
dprg-1 disk xfers: 1170224 9.11/sec 100.00%
dprg-1 drive0: 404842 3.15/sec 34.60%
dprg-1 drive1: 765382 5.96/sec 65.40%
dprg-1 drive2: 0 0.00/sec 0.00%
dprg-1 drive3: 0 0.00/sec 0.00%
dprg-1 total packets: 16391065 127.58/sec 100.00%
dprg-1 input packets: 7683461 59.80/sec 46.88%
dprg-1 output packets: 8707604 67.78/sec 53.12%
dprg-1 input errors: 773 0.01/sec 0.01%
dprg-1 output errors: 365 0.00/sec 0.00%
dprg-1 collisions: 5242 0.04/sec 0.06%

Marke Clinger <>, B.K.Prabhakar <>: OMNI
Solutions in Mt. View runs two NFS benchmarks. One is called "FSBench".
They sell a software tool for monitoring NFS performance called omnimeter.

me: Does anybody seriously believe that you can have half
        as many users banging on a SparcStation and get the
        same response times as with a 4/490?

sid@Think.COM: "Yes. We run 20 diskless nodes on a 4/60 with no problem.
Given that the ethernet is the bottleneck, I would not want to run more
that 40 diskless client on a 4/490."

Hal Stern <halstern@Sun.COM>: Has a "stupid little program i wrote to
measure NFS performance over short periods of time. i wrote it to detect
bursts of requests (over >30/second) on a busy network." He asked me not
to redistribute it, so you can try contacting him directly.

neeraj sangal <matrix!neeraj@uunet.UU.NET>: "We sell a network monitor
(called EtherView) which runs on Sun Workstations. Once of its functions
is NFS Profile...It will tell you the load being handled by each server,
how the load varies over time, what the NFS proc mix (read, write etc.) is,
and the proportion of NFS load generated by clients...& average response
time for each server over all time, over a specific time period and for
each type of NFS proc."

Neeraj Sangal
Matrix Computer Systems, Inc. 7 1/2 Harris Rd, Nashua, NH 03062
uunet!matrix!neeraj (603) 888-7790

Bob Enger <> is co-chair IETF/NOCtools working group
and would like to include any responses on this topic in the NOCtools
catalog of network management and debugging tools. ["The ietf is the
internet engineering task force, a group underneath the internet advisory
board (the old boys that guide the internet). The ietf has working groups
to do different things, like make ospf, snmp, host requirements rfc, etc.
One of their smaller working groups is NOCtools Catalog. We're compiling a
catalog of useful debugging and analysis tools."]

This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:03:55 CDT