MUSBUS

[Ken McDonell]

In article <4034 at phri.UUCP> roy at phri.UUCP (Roy Smith) writes:
>
> 	A computer salesman was talking to me this morning about something
> called MUSBUS.  I vaugely remember hearing about it being some sort of
> benchmark for multi-user Unix systems, simulating some sort of "average"
> multiuser workload.  Anybody know anything about it? ...

MUSBUS (Monash University Software for Benchmarking Unix Systems) is
a synthetic multi-user benchmark in which the user workload profile
is intended to be one of the input parameters.  It was originally designed
to assist in comparative performance evaluations during equipment
acquistions.

Here is the standard "glossy" availability blurb I send out ...

    The MUSBUS source is in the public domain, distributed via the
    USENET newsgroup comp.sources.unix as follows
	    Version 5.0	volume 11, issues 29-32
	    Version 5.2	volume 12, issues 72-74
    You will need both, as the 5.2 distribution is an upgrade kit that
    assumes you already have Version 5.0.

    There is a MUSBUS mailing list that is used to distributed an infrequent
    Newsletter containing items of interest to those using MUSBUS, or related
    performance analysis tools.  If you'd like to join the mailing list
    send an e-mail request to kenj at yarra.oz.au

I am currently working on Version 6.1 with some scaling and normalization
enhancements, and additional "canned" workloads.

> ... Is there some
> repository of MUSBUS ratings for various machines somewhere?

Yes, I have them.  I do not distribute them, because

1. Unlike some others (dhrystone, AIM User Rating, ...) I have no faith
   in ``single figure of merit'' numbers for serious performance analysis.

2. The default workload will help you select the correct machine for Monash's
   Computer Science Department, circa 1982.  The relevance for others is
   questionable.

3. Unlike the pedlars of some other commercially available performance
   numbers, I am not interested in old numbers.  Today's Unix version and
   C compiler will deliver very different performance to yesterday's,
   even from the same vendor and on the same hardware (R&D folks do address
   performance issues in shipped product!).

4. If you are interested in multi-user performance for a particular
   application profile, build a workload and ask the vendors to run MUSBUS
   using YOUR workload.  Software engineering and portability issues have
   been addressed to the extent that this is not an onerous request.
   
The SPEC members have expressed some interest in MUSBUS, anyone from
there care to comment?

Disclaimer: I now work for Pyramid Technology, not Monash University, but
	    as always I speak for myself alone.