jdc@naucse.cse.nau.edu (John Campbell) (12/14/90)
Here's a question that probably doesn't fit into a common Usenet
category. (In other words, forgive me if I picked a bad newsgroup.)
At this site, we have over a thousand PCs, about 400 or so MacIntoshes, as
well as one VMS, one CMS, and one MVS machine. In addition we're
accumulating unix minicomputers (about 4 now) and unix workstations
(about 10 now). Also, we've justed started a campus wide network
using TCP/IP as the main protocol.
The question:
"Which applications should run on which platforms?"
The assumption here is that it may be cost effective to exclude some
applications from some platforms.
Now I'm not often stumped by questions from our director, and despite
the fact the question comes on the heels of a request to acquire a
CD-ROM full of VMS software, I'm intrigued to find I don't have a ready
answer.
My manager found an ACM article (May 1989 Vol 32 No 8) titled
"Computer Hardware Performance: Production and Cost Function Analysis"
by Young Moo Kang. Young shows a scheme for formulating computer
performance as the product of computer characteristics. He maximizes the
profit subject to the performance constraint (and assumes cost is
weighted sum of the performance characteristics) in order to present a
mathematical model that he can apply to leased mainframes. His conclusion
is that there is "no economy of scale for the mainframes and superminis
during the 1981-1985 period."
Of course, we're not concerned here with leased mainframes, but rather a
*very* heterogenous mix of compute power. Also, it's no longer 1985.
We're just now starting to network. Hence our need for sharing data
isn't as great or as obvious as it would be at other sites. One
administrator commented a while ago that he "couldn't imagine a need
for most people to work on anything but a PC".
These are related sub-questions that I wish I knew more about:
1) Is there a clear difference between multi-user and single-user
applications? Is it true some applications should not be done
on a multi-user machine?
2) Is there a measurable cost to the user in starting a task on the
wrong platform? How do data incompatibilities affect work throughput?
3) How do you distinguish between the roles of various platforms?
Why can't you just have lots of PCs and network them if necessary?
4) Why are there files servers and compute servers? What type of
applications/activities really need these devices?
5) Are DEC's NAS, IBM's SNA or HP's NCS (network service brokers) a
realistic answer to the question? When is a network service broker
a detriment (assuming that one was even available)?
Any comments?
Please send me any thoughts you might have. If there is interest I'll
summarize the responses.
--
John Campbell jdc@naucse.cse.nau.edu
CAMPBELL@NAUVAX.bitnet
unix? Sure send me a dozen, all different colors.