Path: utzoo!utgpu!jarvis.csri.toronto.edu!rutgers!usc!apple!bloom-beacon!bu-cs!bzs
From: bzs@bu-cs.BU.EDU (Barry Shein)
Newsgroups: comp.arch
Subject: Re: What is a Mainframe?
Message-ID: <33942@bu-cs.BU.EDU>
Date: 27 Jun 89 23:42:29 GMT
References: <125@ssp1.idca.tds.philips.nl> <20752@winchester.mips.COM> <4400@ficc.uu.net> <187@uvacs.cs.Virginia.EDU> <355@torsqnt.UUCP>
Distribution: comp.arch
Organization: Boston U. Comp. Sci.
Lines: 61
In-reply-to: david@torsqnt.UUCP's message of 26 Jun 89 18:03:57 GMT


Mainframes are generally at the top of the line in overall I/O
(particularly disk) performance, that's the critical measure.

For example, top-end mainframes have separate I/O processors on their
disk channels with significant processing capablities (like the
ability to search for keys independent of the main cpu.)

They also have multi-ported memory systems so many channels can be
pumping data simultaneously. This last fact is critical, a mini or
micro might claim similar I/O performance on a single stream of data
but invariably they can't handle multiple data streams simultaneously
so overall system performance is a fraction of a true mainframe. It's
not unusual for mainframes to be able to pump 16 or more separate disk
channels simultaneously (ie. not by competing for one bus.)

Watching a true mainframe spinning several 200 ips tape drives
simultaneously at full speed (hint: the tapes are spinning at what
looks like full rewind speeds) while keeping a disk farm busy gives
one the right idea of what true mainframe users are after. Most minis
and all micros I've ever seen can hardly spin data between a disk and
a tape without the tape visibly hesitating, particularly when writing
the disk.

I/O capacity is important also, these days mainframes often manage on
the order of a terabyte or more of disk AND CAN ACTUALLY DO SOMETHING
WITH IT IN A REASONABLE AMOUNT OF TIME.

Another common feature is system hardware robustness, it was putting
this type of robustness into minis that earned Tandem its fortunes.

Mainframe users demand a lot of uptime, these systems can't go down
just because a memory board or disk died. It's not uncommon to have
1000 or more users logged in; you can't idle them every time the
system hiccoughs, that's the kind of thing people pay millions of
dollars for and can justify in a large organization.

Minis and micros are not as well defined but (good) minis often offer
mainframe-like performance on a very small number of jobs, you can
move a small group off the mainframe onto the mini without much loss
of performance. The term "micro" generally refers to a certain type of
packaging and manufacturing approach that minimizes costs firstly and
might (these days) give near-mainframe performance on very simple,
single-threaded applications.

But I/O is the real point, 16 simultaneous channels pumping a terabyte
of disk farm while 200 ips tapes spin freely and 1000 or more users
are getting good interactive response...that's the idea and it takes
serious system architectural features to acheive that, you don't do
that with one board or simplistic bus designs (at least not yet.)

Maybe that's a way of differentiating them, micros often have no real
bus, minis usually have one or maybe a very few, mainframes have many
with independent I/O processors and separate paths to/from memory.
-- 
	-Barry Shein

Software Tool & Die, Purveyors to the Trade
1330 Beacon Street, Brookline, MA 02146, (617) 739-0202
Internet: bzs@skuld.std.com
UUCP:     encore!xylogics!skuld!bzs or uunet!skuld!bzs