Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!watmath!clyde!rutgers!mit-eddie!genrad!decvax!tektronix!uw-beaver!cornell!rochester!ur-tut!tuba
From: tuba@ur-tut.UUCP
Newsgroups: comp.arch
Subject: Re: MIPS to offer COBOL
Message-ID: <988@ur-tut.UUCP>
Date: Mon, 26-Jan-87 19:41:27 EST
Article-I.D.: ur-tut.988
Posted: Mon Jan 26 19:41:27 1987
Date-Received: Wed, 28-Jan-87 07:33:07 EST
References: <14392@amdcad.UUCP> <1275@cbmvax.cbmvax.cbm.UUCP> <984@ur-tut.UUCP> <24167@rochester.ARPA>
Reply-To: tuba@ur-tut.UUCP (Jon Krueger)
Organization: Univ. of Rochester Computing Center
Lines: 63

I wrote:
>> For my usual rates, I'll code you a picture(3) library that will look
>> and feel like the real thing, interface to your favorite 4GL, and run
>> at least as fast on a Unix-based 68000 as the COBOL code on an IBM
>> machine costing 10 times as much to buy, operate, and waste programmer
>> time on.  Or has somebody already done this?

dibble@rochester.ARPA (Peter C. Dibble) replies:
>That would be very impressive!  IBM 370 machines have hardware assists for
>formatting numeric output and more powerful instructions than
>the 68000 for handling bcd (they call it packed) numbers.

I write:
Yes, as do VAX-11/750's.  Turns out some execute faster in software
emulation on Microvax II's.  RISC proponents like to offer similar
examples of their load-store boxes outperforming CISCs on tasks
microcoded into the CISC instruction set.


>Is the trick here that you are comparing the absolute top-of-the-line
>68000 box to the bottom of the IBM mainframe line?

Well...I'm depending heavily on the true cost of IBM mainframe cycles.
If IBM were to cut prices on its hardware and software, I couldn't
quite do it.  But as costs stand now, I think I can buy a stripped
M68000-based Unix box for $5K to $15K.  Let's assume it will break
irreparably and its company go out of business in 1 year, after the
fashion of those nasty little start-ups.  You may therefore take $50K
to $150K and start shopping IBM.  Include costs of monthly rental of
operating system and tools.  Identify the target hardware with these
constraints, drag out your COBOL code, time its execution, and I'll
take that as a fair number to meet or beat.

>Maybe you have one of the lobotomized distributed computers in mind?

Not sure what's meant here.

>A version of printf could accept pictures and generate the right output,
>but it wouldn't get you the right "look and feel," at least not from
>the programmer's viewpoint.
>
>The COBOL move corresponding command is an important part of the "feel"
>of PIC output.  I used to use it to compile an output line from fields
>out of several records.  I would think it would be difficult to add
>move corresponding to a 4GL with a library.

You may have me here.  Tell you what.  Supposing I ever code a
picture(3), I'll happily submit both environments to the business
oriented programmer, and accept her verdict on whether it feels the
same to her.

>The last thing I want on my Unix machine is a library that supports
>COBOL-style PIC format conversion, but I'd love to know how you'd get
>the function and the speed.  Especially the speed.

Mainly by constraining the IBM side to one costing ONLY ten times as
much.  In a pinch, I'd ask, reasonably I think, that the timing test
be done on execution of a real COBOL program, not a 370 assembler code
executing the IBM instructions directly.

-- 
Jon Krueger   Department of Necessary Evil   University of Rochester
uucp: {seismo, allegra, decvax}!rochester!ur-tut!tuba   BITNET: TUBA@UORDBV