bzs@BU-CS.BU.EDU (Barry Shein) (03/06/88)
The mainframe seems to be getting more threadbare as each year passes. Typical mainframes these days (eg. IBM3090) deliver about 30MIPs per processor, 1..6 processors and large memory systems which remains somewhat, but not very, impressive. Certainly with the advent of 20+MIPs chips from manufacturers like Sun and Motorola it won't be long before those cpu/memory configurations will be available on machines in the $250K or less price range (I beleve Arrete just commited to parallel processing on the SPARC, Encore is certainly moving towards these speeds, others, including MIPS [the company] are moving a hair's breadth away.) Of course, in the mainframe market there's a lot more than mere central processing performance. You have I/O speed which is critical, reliability and a vendor who will back it all up in many different ways. It's on these latter points that a company like IBM wins big. But what percentage of the computing growth needs mainframes anymore? Back in the days of the 370/168 with its 2MIPs and 4MB of memory it seemed like you needed a mainframe to do much anything and a lot of people cling to that mentality. I have no doubt that JC Penney's or MasterCard need these machines, but what about the zillions of other people who bought IBM back when it was the only thing that would handle their few hundred thousand records (perhaps a few dozen megabytes) and don't realize that today you could process that kind of data on a relatively modest processor? I realize that a lot of these folks keep buying these beasts due to their own software investment and you can't really expect a vendor like IBM to maintain a 10-year old system, so roll in the next model, what choice is there. One big backpressure of course is the need for distributed computation, these days every local department expects to be able to have some of their own computes rather than just telecom to the mainframe. A friend in the biz who handles some Wall St accounts claims that the bigger and bigger mainframe approach is collapsing under its own weight. The systems are too complicated, growing them is putting more and more eggs in a bigger and bigger basket, the systems analysts are feeling seriously out of control of their own systems. Does distributing help? It's not clear, but it's there and the current star topology approach (ie. one big mainframe with all the critical data and a bunch of lesser, second-class-citizen local processors) isn't fitting in. Dogs wag tails, tails wag dogs. Even on the academic campus more and more people I speak to seem to be frustrated enough with the centralized approach (even the modest time-sharing systems) to computing that they would rather deal with less than deal with the bureaucracies these large "data centers" inevitably grow to become. -Barry Shein, Boston University
elg@killer.UUCP (Eric Green) (03/07/88)
in article <8803052348.AA21708@bu-cs.bu.edu>, bzs@BU-CS.BU.EDU (Barry Shein) says: > The mainframe seems to be getting more threadbare as each year passes. > Typical mainframes these days (eg. IBM3090) deliver about 30MIPs per > processor, 1..6 processors and large memory systems which remains > somewhat, but not very, impressive. > But what percentage of the computing growth needs mainframes anymore? > people who bought IBM back when it was the only thing that would > handle their few hundred thousand records (perhaps a few dozen > megabytes) and don't realize that today you could process that kind of > data on a relatively modest processor? Well. Let's see. At educational price levels, a IBM 3090 goes for about $6 million. $6,000,000. Assuming base-line Suns, at $10,000 apiece (average cost, after distributing server costs), you could get 600 Suns for that same price. 600 users. Or, if we look at, say, Pyramids, which certainly can handle everything that a IBM mainframe can handle, at least up to 500mb... average cost, maybe $250,000 apiece (this was 4 years ago, mind you). 4 per mil. 24 for 6 mil. Can handle over 720 users that way, at 30 users apiece. But what do you get for your $6M? Well: You get a line editor that imitates a page editor that imitates a screen editor. You get an operating system less sophisticated than CP/M, and much more complex. You get a system that can handle, maybe, 300 users. Frankly, I don't see why anybody would buy an IBM 3090, unless they needed access to absolutely huge databases that exceed the size of most minicomputer disk packs (e.g. the 1 gigabyte supereagles). > Even on the academic campus more and more people I speak to seem to be > frustrated enough with the centralized approach (even the modest > time-sharing systems) to computing that they would rather deal with > less than deal with the bureaucracies these large "data centers" > inevitably grow to become. Most academic campuses today are still stuck in the 60's, as far as computing center policies go. That is, their policies are based upon MINIMIZING use of the computing facilities, because back in the 60's, computing facilities were a rare and expensive resource. Most of them are having extreme difficulties adjusting to times where a $6M mainframe can handle the entire contingent of CS students online all at the same time, especially now that the CS rush of the early 80's (which produced a temporary problem with computer availability at most colleges) is over and CS enrollments have dropped back down to a more reasonable level. With all that equipment bought to handle the rush still around. E.g. what do you do with a practically unused VMS cluster, when you just bought a brand new 3090 and your professors prefer Unix on a bunch of Pyramids and Encores? Most colleges are still set up for handling, "I have 600 CS students clamoring for access to a 120 user system", instead of "I have 600 CS students, and systems capable of handling 600 users". Note that I don't include Crays and other supercomputers as "mainframes", because they're not general-purpose computers. Although I think Crays are pretty neat, and when I'm a multibillionaire I'll have one for every room :-). After all, on what other computer can you say that GNU Emacs is a small process?! -- Eric Lee Green elg@usl.CSNET Snail Mail P.O. Box 92191 {cbosgd,ihnp4}!killer!elg Lafayette, LA 70509 "Human evolution ended when civilization began".
jmckerna@polyslo.UUCP (John L McKernan) (03/08/88)
In article <3614@killer.UUCP> elg@killer.UUCP (Eric Green) writes: >Note that I don't include Crays and other supercomputers as "mainframes", >because they're not general-purpose computers. Although I think Crays are >pretty neat, and when I'm a multibillionaire I'll have one for every room :-). >After all, on what other computer can you say that GNU Emacs is a small >process?! I think a definition of mainframes which excludes "supercomputers" is overly narrow these days. All mainframe really means is maximum computing power with maximum memory. The reason supercomputers are not general purpose is because general purpose uses have no need of a modern supercomputer's power. It's cheaper and cleaner to use a minicomputer as a front end. Since the demand for increased power and memory is essentially infinite forever, the demand for mainframes is essentially infinite forever. I realize that current usage does make a distinction between mainframe and supercomputer, but I don't see any more real meaning in that distinction than the difference between a car and an automobile. Pre-emptive flame strike: Anyone who disagrees with this is being hopelessly dogmatic and bound to a usage with little regard for its meaning. John L. McKernan. Student, Computer Science, Cal Poly S.L.O. ------------------------------------------------------------------------------- Servant's gossip in Aillas's castle: "Have you heard the latest?, GREEN jewelry is the new fad among the ladies of the court."
vespa@ucscb.UCSC.EDU (Adam Alexander Margulies) (03/09/88)
>Pre-emptive flame strike: >Anyone who disagrees with this is being hopelessly dogmatic and bound to a >usage with little regard for its meaning. > > >John L. McKernan. Student, Computer Science, Cal Poly S.L.O. what? WHAT? A supercomper == mainframe!?! HARDLY, Mr. John L. McKernan. Supercomputers are much more computery! Jeesh. Some people.
elg@killer.UUCP (Eric Green) (03/10/88)
in article <1461@polyslo.UUCP>, jmckerna@polyslo.UUCP (John L McKernan) says: > narrow these days. All mainframe really means is maximum computing power with > maximum memory. The reason supercomputers are not general purpose is because > general purpose uses have no need of a modern supercomputer's power. It's > cheaper and cleaner to use a minicomputer as a front end. Since the demand for > increased power and memory is essentially infinite forever, the demand for > mainframes is essentially infinite forever. 2 The reason supercomputers are not general purpose is simply that they aren't. A general purpose mainframe handles hundreds of users doing all sorts of things. A Cray handles maybe 20 users, running huge number-crunching programs. It's like me trying to say that my Colecovision is a computer. After all, it has a microprocessor, right? But, it simply performs a totally different task. A Cray isn't going to handle 300 users running various and sundry, and your typical mainframe isn't going to run math-oriented programs anywhere near as fast as a Cray. And my Colecovision isn't going to challenge the Commodore 64 as king of low-cost computers -- for one thing, it has no keyboard :-). If there is a falling-out in the mainframe market, as users discover they can do things cheaper with smaller computers, the sales of Cray Research won't be hurt one tiny bit.... -- Eric Lee Green elg@usl.CSNET Snail Mail P.O. Box 92191 {cbosgd,ihnp4}!killer!elg Lafayette, LA 70509 "Human evolution ended when civilization began".
hollombe@ttidca.TTI.COM (The Polymath) (03/12/88)
In article <3614@killer.UUCP> elg@killer.UUCP (Eric Green) writes: > ... Frankly, I don't see why anybody would buy an IBM >3090, unless they needed access to absolutely huge databases that exceed the >size of most minicomputer disk packs (e.g. the 1 gigabyte supereagles). I happened to meet the manager of Northrup Corporation's supercomputer resources at a party a few months ago. She told me they use a 3090 as the front-end processor for their Cray (and it can barely keep up). Most super-minis these days have the same 4 gigabyte virtual address space as the mainframes. The main difference seems to be in the raw processing power (mips) that can be brought to bear on a single process, as opposed to multiple users. 600 Suns are fine for 600 users with relatively modest DP requirements, but if you need to get the payroll for 300,000 employees printed out tonight, because payday's tomorrow, you'd better have a machine that can run that one program in the time available. Of course, a hypercube parallel processor could probably do the job as fast as you could feed it the data, but that's another discussion. -- The Polymath (aka: Jerry Hollombe, hollombe@TTI.COM) Illegitimati Nil Citicorp(+)TTI Carborundum 3100 Ocean Park Blvd. (213) 452-9191, x2483 Santa Monica, CA 90405 {csun|philabs|psivax|trwrb}!ttidca!hollombe
larry@tapa.UUCP (Larry Pajakowski) (03/12/88)
The future of large mainframes is I/O in particular disk or DASD in IBM parlance. The place where I work ( a fortune 100 co.) is adding about 10 DASD devices each month. I'm not sure of the size but most likely over 1gb. each. Try running a large room full of 1gb. disks from your micro ! Larry