bzs@ENCORE.COM (Barry Shein) (02/17/89)
Apollo is shipping a 4-processor box (DN10000) which claims around 100 MIPS performance with respectable floating point and I/O. DEC has just announced the PMAX, a desktop machine based on the MIPS chip which does 25,000 dhrystones (about 13MIPS) for $11,900. That's almost twice the speed of DEC's biggest (read: $1M) VAXes. Rumors abound of impending 40-50MIPS announcements for all the major RISC chips. All have vendors building (typically) 4 processor deskside configurations (ie. 150-200MIPS.) All should be available for well under $100K in the next 12-18 months. Meanwhile the big mainframes (multi-million dollar) are puffing along with about 100MIPS in their biggest configurations. What does it all mean? Perhaps a paradigm shift? BIG no longer equals FAST and vice versa? When you add in the fact that the big mainframes typically have 100-250 users log in the implications become breathtaking. We'll leave the exotic machines (eg. Cray) out of this for the moment since they're relatively rare (and the few that exist tend to be overcommitted.) In a way it makes sense, at this point in history what magic would the big machine builders have that would make their CPUs fast? Will it soon be the mark of the quaint dinosaur who looks at a desktop workstation and says "I need a big, fast machine!" Sort of like a person who still thinks data is only safe when it's on punchcards? Or that real programmers use assembler? Sure, there's still I/O. I/O speed will probably fall as a factor which distinguishes big machines from small ones for the same reason...*speed* comes from chips, and chips ain't big! What won't fall so fast will be I/O size. Few people want a terabyte spinning in their office, the sheer size of that kind of equipment remains a limit for the foreseeable future (not that it won't happen, but when data becomes large there's still an economy to sharing it out of a central location, particularly when it's costed on a per-copy basis.) The point is obvious, organizations which have built their empires by providing MIPS are, in most cases, doomed... -Barry Shein, ||Encore||
janssen@titan.sw.mcc.com (Bill Janssen) (02/17/89)
In article <8902162247.AA28231@multimax.encore.com>, bzs@ENCORE (Barry Shein) writes: >The point is obvious, organizations which have built their empires by >providing MIPS are, in most cases, doomed... Most MIS groups have built their empires on accounting and services, not on MIPS. You're still going to need someone to work out the corporate licencing on Lotus 1-2-3, not to mention storing the backup tapes. Bill
kent@WSL.DEC.COM (02/17/89)
I dunno. I have a PMAX (DECStation 3100) in my office. I like it a lot, but it doesn't suffice for all my computing needs. It doesn't have enough memory, and I'm unwilling to have rotating storage in my office (too much noise and heat). To be honest, I don't want the CPU box in my office. I'm constantly amazed at the noise level that people are willing to tolerate in their work environment. I gave up and put a paging disk in, because the increase in performance was enough that I'm willing to put up with the noise. For now. But I still get most of my cycles from a personal 15 MIPS machine with four times the memory of my PMAX and a Gigabyte of disk that sits in a machine room, and I like it that way. I think there's still a place for mainframe-class I/O performance, but not at today's prices ... except in environments where traditional timesharing is going to be in effect. If there was a reasonable X terminal (read 68020-class performance with no memory problems) I'd probably prefer that to a PMAX in my office. (Of course, I work in a hardware-rich research lab. But we're talking ideals here.) Cheers, chris
mwm@VIOLET.BERKELEY.EDU (Mike Meyer, I'll think of something yet) (02/18/89)
>> >The point is obvious, organizations which have built their empires by >> >providing MIPS are, in most cases, doomed... Barry, that's ambiguous. There are two types of such organizations - those who sell Big Iron, and those who administer and run same for some organization. Those in the first group seem to be switching to selling smaller systems quite handily. Those in the second group aren't in quite that bad a shape. Let's look at Universities, because that's what I'm familiar with as large bureaucracies (I tend to work for small firms when I'm in the real world). Right now, computing needs for any given department can be met in any of five different ways: 1) Buy and administer a Unix box. Not many departments have the expertise to do that. 2) Buy a Unix box, and let someone else administer it. Currently, this is slightly cash-intensive for some departments, but for some it works well. 3) Buy and administer one or more PC-class machines. This is becoming increasingly popular. 4) Buy and let someone else administer the PC-class machines. This isn't much used, because the cost of the administer greatly outweighs the cost (and value) of the machine. 5) Rent time on someone elses box. Now, consider how the advent of super-mini & faster small boxes is going to affect each of those options. 1) No change, except they'll get more power. But these folks aren't buying services from the MIPS-providers anyway. 2) Once again, no change - except they'll be able to afford more power. This may encourage some of the MIPS providers to start providing the service of administering machines. 3) Once again - I can't see much of a change, unless those cheap super-minis become as easy to administer as a PC. 4) Here we see a real difference - the value of the machine will go way up, so it will become cost-effective for more people to take that route. This may grow by as much as an order of magnitude; which will still leave it very small. 5) No change, except that "someoene else" gets to buy more MIPS for less, which will hopefully be passed on to the customers. As you can see, the thing that will save the "Computer Center" is going to be the expertise needed to run those super-minis. If they can make the step to providing those services on a box someone else owns (or renting a box to someone else "wholesale"), they will stay in business. As was mentioned by someone else: >> Most MIS groups have built their empires on accounting and services, not >> on MIPS. You're still going to need someone to work out the corporate >> licencing on Lotus 1-2-3, not to mention storing the backup tapes. Those groups can also stay in business by providing backup services for servers, and by doing the paperwork for dealing with site licenses for software. Here at UCB, the computer center is already doing: 1) Wholesale leasing of machines to others. 2) Administering machines for others. 3) Running clusters of workstations. 4) Running servers for clusters that belong to someone else (thus providing them with backup&restore services). 5) Providing software for remote mounting or use to the campus (i.e. - an NNTP server available to anyone on campus). 6) Providing consulting, software and the interface for "university pricing" on PC-class machines. Finally, there are some services that just won't go away. There will be departments that just don't want to - or can't afford - to deal with any machines in their office. They will wish to buy time from someone else, and will buy either time on a workstation cluster, or on a traditional timesharing box. Likewise, there are things that many departments can't afford - like a phototypesetter. And of course, someone has to be responsible for he network that lets people get to all those ervices. Providing these things will continue to stay in the computer center. There is one group I failed to mention - those who need deep MIPS (as opposed to broad MIPS), who are renting from someone else because they can't afford to buy a machine big enough to meet there needs. The new class of super-minis will cause that group to shrink as more can afford something to meet there needs. If you've got a Cray running timesharing with to large a load, then having your own CPU in the 10-50 MIPS range might get you results in real-time faster than running them through the Cray. It's going to be interesting. I don't think any major manufacturers, or any large MIS organizations are going to have serious problems. They may have to reorganize and reorient themselves, but that's about it. <mike P.S. - Yes, I work for just such an MIS organization. But we're reorganizing (far to often), and reorienting to providing slightly different services.