Path: utzoo!mnetor!uunet!lll-winken!lll-lcc!lll-crg.llnl.gov!brooks
From: brooks@lll-crg.llnl.gov (Eugene D. Brooks III)
Newsgroups: comp.arch
Subject: Re: Single tasking the wave of the future?
Message-ID: <1406@lll-lcc.aRpA>
Date: 29 Dec 87 20:06:37 GMT
References: <18@amelia.nas.nasa.gov> <2341@encore.UUCP> <25@amelia.nas.nasa.gov> <1030@alliant.Alliant.COM> <3771@ames.arpa>
Sender: usenet@lll-lcc.aRpA
Reply-To: brooks@lll-crg.llnl.gov.UUCP (Eugene D. Brooks III)
Organization: Lawrence Livermore National Laboratory
Lines: 25
Keywords: parallel processing today

In article <3771@ames.arpa> eugene@pioneer.UUCP (Eugene Miya N.) writes:
>In article <1030@alliant.Alliant.COM> muller@alliant.UUCP (Jim Muller) writes:
>>> >Debugging is a whole other soapbox, but my experience is that debugging
>>   Race conditions are typically a problem only when a compiler has been
>>   inappropriately "permitted" to optimize code that it normally would have
>>   refused.  The obvious approach, of course, should be to debug in single-
>>   processor mode.
Although the above comments are a little off the wall, the debugging of a
parallel program is best done in two steps.  First you run it in
serial mode (parallel code but with one processor) and find and kill
all of the bugs that show up.  If your program dumps core here the
use of a good serial debugger will be of great help.  After doing this
you run the code with more than one processor.  Bad things that are
hard to pin down can still happen, (gee Eugene, this program ran fine
through 13 cycles and then dumped core out of the blue!) but at
least you won't have the serial program bugs biting you when you are
running the code in parallel.

Of course there are some implicit assumptions:
	1) The parallel programming model you use uses the same code
regardless of the number of processors (the number of processors
is a runtime parameter that may not even be known to the user of the
code)
	2) The code produces the same results (perhaps to machine roundoff
accuracy to satisfy those who have real numerical parallel code experience)
regardless of the number of processors you run it with.