Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.3 Unisoft-Cosmos; site kepler.UUCP
Path: utzoo!linus!decvax!ucbvax!ucdavis!lll-crg!well!micropro!kepler!glen
From: glen@kepler.UUCP (Glen Rotan)
Newsgroups: net.arch
Subject: Re: Latest Drystone Benchmarks Posted to net.sources
Message-ID: <302@kepler.UUCP>
Date: Fri, 18-Oct-85 16:18:51 EDT
Article-I.D.: kepler.302
Posted: Fri Oct 18 16:18:51 1985
Date-Received: Mon, 21-Oct-85 05:22:06 EDT
References: <145@vaximile.UUCP>
Reply-To: glen@kepler.UUCP (glen)
Organization: MicroPro Int'l Corp., San Rafael, CA
Lines: 23

There have been several postings lately of timings with
the dhrystone benchmark.  It is important to note that the
the timings for UNIX systems were done using the time(2) function.
This function measures total elapse time rather than the time the
computer spends on this program.  The times given by this method
represent not only the speed of the computer system, but also how
busy it is running other jobs.  Using this method the system I'm
on yields results ranging from 107 to 352 dhrystones per second for
the non-register version.  (frank@sagan.uucp got 305 as the highest
on a twin to this one.  I think I got the larger number because
I was a little more persistent).

There are two ways around this.  The easiest is to take timing out of the
program and use the time(1) command to measure the time spent on this task.
The other way is to modify the program to use the times(x) function to
do timing.  Both give 264 and 290 dhrystones per second on the Cosmos
for non-register and register runs respectively.  The deviation using either
of these has not been more than +/-10 dhrystones.
--
glen

Glen Rotan @ MicroPro Product Development
{hplabs,dual,ptsfa,lll-crg,glacier}!well!micropro!kepler!glen