Unix on IBM Mainframes & Clones

[Dave Tweten]

From: bjorn at alberta.UUCP (Bjorn R. Bjornsson)

	I would like to here about peoples experiences with
	Amdahl UTS and/or IX/370 under VM.  Indications of
	implementation quality, efficiency, horror stories,
	price (including maintainance if available), etc.,
	is what I'm looking for.

We have been running Amdahl UTS/V version 1.0, under VM version 3, on two
Amdahl 5840s for about a year.  We just got the 1.1 and 1.1.1 UTS/V updates,
but will probably hold out for UTS/580 (which is on order).  We are also
about to upgrade to VM/HPO version 4.  Both flavors of UTS are basically
System V.2.

The good news is that UTS supports a lot of users and a lot of 3380 look-alike
disk.  Up to 100 9600 baud RS-232 connections at a time can be made through our
Micom switch and Amdahl 4705E communications controller.  We support 50
simultaneous production users, or so, for most of the day, on one virtual
machine, on one of the 5840s.  We are developing one of the 5840s as a GIANT
disk server (about 40 Gig, to start) for our ethernet and HyperChannel based
TCP/IP networks, which include a Cray-2.

There is also some bad news:

1.	The TCP/IP support on the ethernet had to be added-in (Wollongong), and
	we had to roll our own HyperChannel support in UTS (not strictly
	necessary anymore) and in VM, to permit virtual machines to share
	adaptors (still required).

2.	Version 1.0 of UTS had many bugs (so what do you expect from version
	1.0?).  We did a LOT of bug fixing and Amdahl fix integrating.
	Supposedly, version 1.1.1 is much better, but we'll probably never
	know.

3.	The C compiler occasionally produces some strange code.  For a while,
	that prevented Unipress EMACS from working in all its glory.  It STILL
	doesn't work in all its glory, though the latest reason is unknown.

4.	Because UTS/V depends upon VM, it is unprepared to deal with all the
	perversity of I/O devices in the real world.  That can be the source
	of some considerable headaches if you want to roll your own drivers for
	bizzarre devices (as we did for the HyperChannel).  Again, the
	advertizements say UTS/580 will make it all better.  We'll see.

5.	High speed disk I/O through UTS/V is a problem.  It seems to do only
	sector I/O.  The ability to support track I/O would push maximum
	sustained I/O rates above their present 2 megabit per second upper
	limit.  So would cached controllers (for read), or a ramdisk (for the
	truely desparate).

On balance, I think we are reasonably satisfied, all but the if-you-have-
EMACS-you-don't-need-a-shell crowd.