[comp.sources.wanted] benchmarks

terry@pride38.UUCP (sys mngr) (02/01/90)

here's something we all want but just haven't asked for;
a benchmark program for Unix. A program that stands alone on a floppy!
just take your copy in the store and get some results that are useful to
multi-users like this would test disk access, cpu speed, bus speed and what ever
else you could get on a 5 1/4 disk....

terry @ ..!pride386

tinle@aimt.UU.NET (Tin Le) (02/02/90)

In article <185@pride38.UUCP>, terry@pride38.UUCP (sys mngr) writes:
> here's something we all want but just haven't asked for;
> a benchmark program for Unix. A program that stands alone on a floppy!
> just take your copy in the store and get some results that are useful to
> multi-users like this would test disk access, cpu speed, bus speed and what ever
> else you could get on a 5 1/4 disk....
> 
> terry @ ..!pride386


I agree with you about the need for a minibenchmark that can help a
user considering the purchase of a UN*X system.  It won't be a thorough
benchmark (not supposed to be), but will be useful as a first cut type
to eliminate systems you don't want to waste time on.

AIM Technology sells a series of UN*X benchmarks.  Our benchmarks are
meant to be used by OEMs and manufacturers (they all come with source
license).  I've started putting together a simple minibenchmark using
bits and pieces of our current benchmarks.  Even got it working to
generate a series of simple numbers (similar to the DOS SI ratings, or
the Landmark numbers).  Of course, these numbers are about as meaningful
as the "MIPS" number (Meaningless Indicator of Performance Statistics).

They are useful only as a relative ranking to remove systems from
further consideration.

The minor problem is that the minibenchmark I wrote compose of source
code fragments from our other products and can not be release into the
public.  I've thought about this for a while and came up with several
possible solutions:

1. UN*X in the 80x86 world is/will be standardized enough to actually
   have binary portability (similar to DOS programs).  I can release
   binaries that will run under S5R3 compiled for a 286 that is
   guarantee to run on most 80x86 based systems running Sys5 compat
   OS.  I am assuming that no one wants to buy an 8088/86 system to
   run UN*X on (do you know of any?).

2. The same can be said for binary ABI for the Sparc, MIPS, and 88K.

Voila, no need to release source code.  The last hurdle now is that
I am working on a new project and have not had time to work on this
minibenchmark.  If there is enough interest (read: financially self
supporting), I can try to convince management into letting me work
on this and releasing it.

Here are details on the minibenchmark.  It will give numbers for 7
categories:

	- System calls
	- Function calls
	- Integer Math
	- Floating Math
	- IPC (shared-mem, mesg q, pipes) throughput
	- Disk r/w/copy throughput
	- RAM throughput

It will do some 'magic' with the 7 numbers to output a single number
representing the machine's relative ranking to some standard system.
I decided on using a plain 80386/16Mhz/4MB RAM/120MB ESDI HD system
as the standard (because that seems to be a common configuration).

I have a single binary that will run on Xenix 286/386, ISC 386,
Microport 286/386.  How many of you out there are really, really
interested in something like this?  (A common minibenchmark).

-- Tin Le