[comp.unix.questions] Roles for mainframe unix and UTS vs. AIX/370 questions

DMOYNIHA%WAYNEST1.BITNET@cornellc.cit.corn (Dennis P. Moynihan) (09/30/89)

We're in the process of reviewing mainframe based Unix systems to
run on our Amdahl 5890 (IBM compatible) processor.  We're looking at
bringing in Amdahl's UTS Unix and are reviewing AIX/370.  Our central
facility is not currently a unix shop, so I've got a few questions for
the group:

1.  What are valid roles for a mainframe based unix system?  Obviously
    departments have access to increasingly powerful workstations--we're
    looking for the right niches for each class machine.  Right now
    we see a mainframe unix system as a large file server, good for long
    running processor intensive tasks, and for handling large number of
    trival-task users (i.e. undergrads) cost effectively.  Are these valid?
    Others that we missed?
2.  For those that have selected or are thinking about selecting UTS or
    AIX/370, what prompted your decision?  What are the general advantages
    and disadvantages of each (in your view)?
3.  One of our concerns is application integration and porting between
    our departmental processors (mostly Ultrix and SUNs) and our mainframe
    system.  UTS is in lockstep with AT&T developments, while IBM seems
    to be claiming that AIX/370 will have the "best" of both System V
    and BSD.  What are your experiences with or thoughts on porting to
    these systems?

We're open to any other general comments you have on mainframe unix
systems and UTS and AIX/370 in particular.  Thanks in advance!

--------------------------------------
Dennis Moynihan    (DMOYNIHA@WAYNEST1)
Computing and Information Technology
Wayne State University
Detroit, MI

jsloan@handies.ucar.edu (John Sloan,8292,X1243,ML44E) (09/30/89)

From article <21015@adm.BRL.MIL>, by DMOYNIHA%WAYNEST1.BITNET@cornellc.cit.corn (Dennis P. Moynihan):
> We're open to any other general comments you have on mainframe unix
> systems and UTS and AIX/370 in particular.  Thanks in advance!

I don't have any experience specifically with either UTS or AIX, but I
do have a few general comments (although from your questions I bet
you've already considered them). My perspective is that I spent years
as a systems programmer on IBM mainframes (360/65 and 370/3031) under a
variety of OSs (OS/MFT, MVT, SVS, MVS); for the past five or six years
I've been dealing almost exclusively with UNIX (both BSD and System
V).

I personally know of two AIX/370 shops, who chose AIX just because they
had IBM hardware and hence had IBM support.  You may end up using UTS
for the same reason.

One thing IBM/compatible mainframes are really good at is providing I/O
bandwidth, which makes them REALLY attractive as file servers. You may
want to consider though, that much of that bandwidth exists in the
communication paths between CPUs, channels, and controllers. Once you
leave this optimized I/O path, you may find that the channel and/or
network speed is not all that fast in comparison.  If you use the
mainframe as a fileserver over traditional access paths (i.e.
Ethernet) then it may turn out you've got this I/O path the width of
the Mississippi, all funneling through 2 inch diameter pipe.

In those hazy times when I thought about using a mainframe as a file
server, I've wondered if there was hardware to channel attach a
FDDI-based fiber network; if not, then a way to channel attach multiple
Ethernet interfaces, making each interface look to the outside world as
if it were attached to a separate file server (different Internet host
ids, etc.). I'd probably like to spread those Ethernet interfaces
across several channels, too. Depending on the statistical distribution
of channel usage for disks, tapes, Ethernet, etc. I might want to try
to optimize the combinations (if possible).

Finally, do you really need to run UNIX at all? Some possibilities:

o	Run AIX/UTS as the native operating system.

o	Run AIX/UTS under VM as a guest operating system.

o	Run AIX/UTS under PR/SM.

o	Run TCP/IP+NFS on top of MVS.

This latter alternative doesn't give you a big UNIX timesharing engine,
but it retains MVS functionality, without the overhead of VM or PR/SM,
and could still let you use the big iron as a file server (no claims
are made, though, about mapping UNIX file system semantics on top of an
MVS VSAM-based file system).

There are also network bandwidth/management issues regarding
centralizing a heavily-used file server resource; some of those same
issues crop up when you just put all your Sun (or whatever) file
servers off the same network cable (e.g. together in a computer
room).

As you come to conclusions, you would be doing the net a service
by posting a summary. I for one would be really interested in
hearing what you've decided, how you came to your decision, and
how it all works out in the end. I hope that eventually EDUCOM
publishes another volume similar to their ancient _Campus Computing
Strategies_ and their more recent, excellent, _Campus Networking
Strategies_, which covers some of these very issues with case
studies. Maybe a _Campus Integration Strategies_.

John Sloan            NCAR/SCD             NSFnet: jsloan@ncar.ucar.edu
P.O. Box 27588        P.O. Box 3000        NCAR Mesa Lab, Room 42A
Lakewood CO 80227     Boulder CO 80307     +1 303 497 1243
Logical Disclaimer: belong(opinions,jsloan). belong(opinions,_):-!,fail.

KJH%APLVM.BITNET@cornellc.cit.cornell.edu (Kenneth J. Heeres) (09/30/89)

We had UTS installed at our system and have removed it.  We are very much
interested in mainframe UNIX but see it in a role as server (file, database,
compute, etc).  UTS had very poor network support and good 3270 terminal
support.  UTS/580 appears to be much better at network support.  AIX has no
3270 terminal support but it does appear to have good network support.  We
would love to have either AIX/370 or UTS/580 in house.  Unfortunately the
fees are too much for us to justify.  We are going to experiment  with
TCP/IP under MVS and VM instead.  I would be very interested in any
comments you get back.

                      Kenneth J. Heeres
                      The Johns Hopkins University
                      Applied Physics Laboratory