Path: utzoo!mnetor!uunet!husc6!think!ames!ucsd!sdcsvax!ucsdhub!hp-sdd!hplabs!hp-pcd!uoregon!dboyes
From: dboyes@uoregon.UUCP (David Boyes)
Newsgroups: comp.os.vms
Subject: Re: Help us defend against VMS!
Message-ID: <1648@uoregon.UUCP>
Date: 5 Mar 88 20:32:04 GMT
References: <2235@bsu-cs.UUCP> <892@cfa.cfa.harvard.EDU> <20403@bu-cs.BU.EDU>
Reply-To: dboyes@drizzle.UUCP (David Boyes)
Organization: University of Oregon, Computer Science, Eugene OR
Lines: 58

In article <20403@bu-cs.BU.EDU> bzs@bu-cs.BU.EDU (Barry Shein) writes:
>
>From Steve Ward:
>>VMS has an excellent Fortran compiler and any serious Fortran number
>>cruncher in research will cry its virtues.  Unix only has one Fortran
>>compiler{  in the same league with the VMS compiler, and it is the DEC
>>Fortran-for-Ultrix compiler.

Steve has a good point here. VMS Fortran isn't bad -- it has a lot of
useful extensions and in general could almost convince me to go back
to writing Fortran code. DEC put a lot of time into making this onw
work -- would that they had spent as much time on the rest of their
language compilers (you aborters of VAX COBOL listening out there? --
get it right, you .... well, nevermind.).

My major point of issue with VMS Fortran is that 1) it is non-standard
in a lot of ways -- those useful extensions just don't move well, and
2) no machine that it runs on is fast enough to do serious number
crunching. I've been processing matrixes of dimensions larger than
1,000,000 X 1,000,000 elements as part of an image processing system,
and even an empty 8800 just doesn't cut the mustard to handle that
kind of data manipulation. We're talking DAYS of compute time here, folks.

> even DEC's largest machine can't be considered a serious
>number cruncher's machine in this day and age,

This is very true. DEC highly overrates the compute power of their
hardware. I find it extremely interesting that an IBM 4341-II (about a
780-class machine) can manage to do a comparable job to the 8800 I
mentioned above -- within 3-5 hours of compute time difference. DEC's
got to be doing *something* wrong here.

Someone once told me that 1 IBM MIP was roughly equivalent to 2.5 DEC
MIPs. Anybody else heard this and want to confirm?

>In fact, I think DEC has done an unfortunate disservice, more of a
>sociological phenomena, by plying to the scientific community and then
>leaving them out in the cold with incompatible fortran

In all fairness to DEC, you don't have to use the extensions at all
and things will port very well. I move programs from VMS Fortran to VS
Fortran on our IBM box all the time.

>and no iron to
>grow into and remain competitive with their more nimble colleagues.

True. I'd kill for a medium sized Masscomp or a Convex. Even a medium
size IBM 9370 would be an improvement.

>
>	-Barry Shein, Boston University


-- 
David Boyes         | ARPA: 556%OREGON1.BITNET@CUNYVM.CUNY.EDU
Systems Division    | BITNET: 556@OREGON1
UO Computing Center | UUCP: dboyes@uoregon.UUCP
'How long d'ya think it'll be before just us oldtimers remember WISCVM?'