Path: gmdzi!unido!mcsun!uunet!stanford.edu!ames!sparkyfs.erg.sri.com!zwicky
From: zwi...@erg.sri.com (Elizabeth Zwicky)
Newsgroups: comp.org.usenix,comp.unix.admin,comp.unix.large
Subject: LISA V transcript: Alternate track on Dumping the Mainframe
without Crushing the Users
Message-ID: <1991Oct18.202053.4988@erg.sri.com>
Date: 18 Oct 91 20:20:53 GMT
Sender: ne...@erg.sri.com
Organization: SRI International, Menlo Park, CA
Lines: 101
Xref: gmdzi comp.org.usenix:2015 comp.unix.admin:43079 comp.unix.large:343
Originator: zwi...@sparkyfs.erg.sri.com
The notes from the alternate track session entitled "Dumping the
Mainframe without Crushing the Users" follow. These notes were taken
by Paul Hilchey, hil...@ucs.ubc.ca. The file name is "dumpmain-alt"
This is a transcript of a session at the LISA V conference in San
Diego. It is not a verbatim transcript of what happened; these
transcripts are amateur efforts, done by various people with a wide
range of attempts at accuracy. No speaker is representing the company
or companies for which he or she works, unless otherwise noted.
Corrections are welcome, and should be mailed to zwi...@erg.sri.com.
In particular, please let me know if you recognize yourself as one of
the many unidentified speakers. These transcripts, and other items
relevant to the conference (copies of overheads, papers, and programs)
are available for anonymous ftp on ftp.erg.sri.com in ~ftp/pub/lisaV.
That archive is growing as I finish off transcripts and transfer data.
It will be shadowed at other sites, to be announced later.
Dumping the Mainframe without Crushing the Users
Peter Van Epp, Simon Fraser University
Ian Reddy, Simon Fraser University
This session was a discussion of experiences with migrating from
a central mainframe to a distributed computing environment. Approximately
25 people attended.
To start things off, Peter Van Epp and Ian Reddy recounted their
experiences at Simon Fraser University. Their academic mainframe environment
is an IBM 3081 serving 200 simultaneous users and providing Bitnet and
Datapac (public X.25 network) connectivity. There is also a tape library
of about 10,000 IBM 3420 tapes, and a Xerox 4090 laser printer.
Dissatisfaction with the mainframe stemmed mainly from slow e-mail
performance, and the university was unwilling to upgrade the hardware
due to high costs. Instead, a committee of academics was appointed to
select replacement computing equipment. The initial plan was to take
the $300,000/year mainframe maintenance costs and buy Sun
SparcStations for each department. This plan was abandoned when the
committee found out what the software licensing fees for products such
as SAS would be for that many machines. In the end they purchased
- an Auspex file server with 10 gigabytes of disk
- 4 Silicon Graphics 4D/320 servers (1 general use, 3 for research)
- 2 Sun 4/470 servers (for instructional use)
The Auspex was chosen as it was the only product that could offer anything
close to the disk I/O bandwidth of the mainframe. By keeping all home
directories on one file server, backups are made manageable.
The equipment arrived in late August, less than a week before the deadline
for having e-mail off of the mainframe. After much intense activity,
the service was first made available about 10 days after the deadline.
They now have about 20,000 accounts on the system and this has caused some
problems:
o It was taking about 7 minutes to build the NIS password map and
a further 7 minutes to push it to one slave server. This was
alleviated by replacing the map building shell script with a
C program, and by updating the map every 30 minutes instead
of every time a password is changed. Others reported that
they were using rdist instead of NIS.
o Sendmail is some situations walks through the entire passwd.byname
map to search the gecos fields. They have not found a solution
to this yet.
o NeXT computers copy all of the NIS maps every 30 minutes, which
jams the network. They have not yet found a permanent way to
disable this.
o Having 15,000 undergraduate home directories in one directory
proved to be too much for Sun's File Manager, and also causes
performance problems for NeXTs.
Printing is being done on 6 HP LaserJet III SIs that are attached to
a Novel network, and charged for with laser cards. They also plan
on using their Xerox 4090 once the PostScript support for it works
well enough.
Elm was chosen as the e-mail interface on the Unix service, with
a modified POP-mail for PCs and Eudora for Macintoshes. They would
have preferred to use an X.400 mail system, but could not find any
suitable user interfaces. One major deficiency with Unix mail is the
lack of user creatable groups, something that was a popular feature on
their mainframe. There followed some discussion about using X.500
to provide these services.
Peter commented that he would rather have waited before embarking onto
the distributed computing path, or have used MIT's Athena software on
DEC computers. He added that they had not yet had enough experience to
know if the hardware they chose will be adequate, but that they can
expand the SGI machines with more CPUs if needed. Several people said
that they want to hear from him again in six months to see how they are
making out.
A general observation was made that changes of this sort are particularly
difficult and expensive because they involve two transitions: from a
mainframe to workstations; and from a mainframe operating system to Unix.