SUMMARY: Analyzing a Sparc Network

Date: Thu Aug 27 1992 - 08:32:34 CDT

Received: by opus.isdl (4.1/SMI-4.0) id AA06167; Wed, 26 Aug 92 11:49:31 EDT

Sun Managers,

Thanks for the input. The original message follows.

> We have a network of about 50 sparc (2s, ipx's ) and we have 3 sparc 2
> based file servers. Most of our users have about a 1/2 a gig of
> storage on their local machine. Shared files and applications go on
> the servers.

> The question is I want to be able to analyize the performance and load
> of the network, the fileservers and any other "Key" machines that I
> might deem necessary. The purpose is to be able to determine if one
> server or workstation is being overused or overwhelmed by data or
> processing. This will enable us to move applications and shared files
> to help balance the load on the network. (It will also give me
> something I can wave in front of management's face to help get me newer
> machines.)

> One solution was Sun's sunnet manager, but I almost lost it when they
> gave me the quote. It is awefully expensive. And it didn't seem to
> have any analysis tools included...Just data gathering.

> Another solution was to use sun's perfmeter tools (maybe netstat &
> nfsstat) and write some postprocessing tools to allow me to graph load,
> utilization, packets, etc... As well as perform some meaningful
> statistical analysis.

> This was the solution I was planning to take, but I figured I would see
> if anyone else out there has any suggestions or programs (reasonably
> priced programs).

Thanks to:

My feeling is that there a lot of rich companies out there, and a lot of poor
ones, but not very many in between. Many of the solutions were netmetrix or
other similar programs that cost $$$$. Perhaps as we grow larger I will be
able to justify such a tool, but not at the moment. Others pointed me at
papers and public domain software (now we are in my price range :-), but I
haven't had an oportunity to follow them up in detail yet. One person told me
that this was the wrong place to post this -- Sorry if that is so.

I think the solution is going to be a collection of home grown perl/awk/csh/C
scripts that we write over the next couple of months. If anyone out there has
any new suggestions or never replied to my original post, please drop me a

Michael Kirby
Image Systems Development Laboratory


NetMetrix Release 2.60

New Features:

        Completely revised Load Monitor and NFS monitor!

                * New Graphics and Charts.
                    - Pie charts, Histograms, Line graphs.
                    - Adjustable scaling

                * Continuous Independent Segment Monitors.
                    - Start and stop at the command line.
                    - Multiple displays may be attatched to the segment
                      monitors from anywhere on the internet.

                * Run Segment monitors without requiring a display.

                * Much more efficient memory architecture.

                * Revised ZOOM Architecture no longer requires ZOOM
                  configuration prior to data collection.
                    - ZOOM paths constantly available
                    - Full ZOOM structure can be instantly displayed on
                      screen with 4 levels of ZOOM.
                    - Change any level or selected item and see the ZOOM
                      path change instantly.

                * Monitors capable of dumping data to disk allowing
                  extensive, long-term analysis of network activity.

Call Metrix at 1-603-888-7000 for other info.

Btw, are you aware of a powerful network monitor software called
Etherview or Netscope from Qualix group?

It compiled for Openwin and from what i've heard, it was pretty impressive.
Qualix number is 1-800-245-UNIX or (415) 572-0200 email:


One of our divisions uses SharpShooter (from AIM) and NetScope, and swear
by them. Both of these are marketed by Qualix. They ain't cheap, though.

We don't use them here because there are certain circumstances under which
one or both don't work (like, on an Auspex), but I have given them a test
drive and the X interfaces are very pleasant to work with.


You might want to look at a paper presented at the last USENIX LISA conference
by Carl Shipley and Chigyow Wang entitled, "Monitoring Activity on a large Unix
Network with perl and Syslogd." The article asks you to,"contact the authors to
get copies of the meter program and example perl scripts for use in system
monitoring." Their email addresses are:


Check out SYSMON from BALR systems.

They provide a system which is relatively inexpensive (~10k for a
global site license) and they give you the source code. It does both
real time and historical statistics for Sun machines and will display
them using 2D and 3D graphics.

Send e-mail to '' to get more information.


look at rfc1147 (anonymous ftp from .... its kinda big but
lists lots of tools that are freeware and *might* address your needs.


A Network General Sniffer will do what you want (but they are not
inexpensive). They make portable ones that you can transport to and
monitor/analyze a particular network, or they have some which you
install (one per subnet) and then access remotely over the net (or a
separate management subnet) [This is their DSS line; soon they will
even have the DSS stuff accessible from X on a SUN].

Again, not inexpensive, but a good solution.


Look at tcpdump 2.2...It's done by Van Jacobsen at Lawrence Livermore, one
of the world's leading authorities on tcp...its an evolution of etherstat(8).
Machines are moving around so I don't run an anonymous ftp site within Xerox --
look on archie for where it is and get a new copy off the internet.

Read some of the ethernet papers (Boggs '88 paper from decwrl comes to mind).

Ever since my Alto days, people blamed the ethernet for latency. Ethernet
is truly a wonderful medium...Don't guess...You need facts and figures to
support hunches...

Have you looked into FDDinet (10x ethernet). Generally, as processors got
faster, networking takes up more and more time...What was acceptable with
68xxx may not be acceptable with sparcs...

[This is my favorite solution...Go so fast so you don't need to measure :-) ]


This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:06:48 CDT