rdt@houxk.UUCP (R.TRAUBEN) (04/23/84)
RISC is a neat idea but it has gotten enormous air play. It is no longer risque' enough for my taste and often borders on trite. There must be other architectural concepts we can kick around. May I suggest a few to get the ball rolling? 1. how do supercomputer vectoring compilers work. 2. OS issues of mutiple cpus trying to run different parts of the same process. 3. solutions to the multiple cache, multiple copy, coherency problem? rdt
nather@utastro.UUCP (Ed Nather) (04/24/84)
< - Take that! ->
A question about current architectural research:
There seem to be two currently popular basic configurations for
computer systems: a) Time-shared cpus that run many jobs "simultaneously"
on a single cpu (or a pair of them); b) single- user mini or micro computers.
Is there any economy of scale in building a set of independent cpus in one
box, such that each user gets one of them, sharing memory and i/o resources?
The memory would have to be multi-ported -- is this impractical?
--
Ed Nather
ihnp4!{ut-sally,kpno}!utastro!nather
Astronomy Dept., U. of Texas, Austinsmith@umn-cs.UUCP (Richard Smith) (04/25/84)
#R:houxk:-25300:umn-cs:13900009:000:801 umn-cs!smith Apr 25 18:48:00 1984 [] Another interesting 'problem' might be those in trying to implement co-processor systems. It seems to me that 'old software' is a key problem that newer architectures need to face: people are less and less apt to discard old, working software when new hardware appears [case in point: BBN designed the C30 processor to perfectly emulate the crufty old Honeywell 316s that the Arpanet IMP software runs on since 316s haven't been sold for about a decade. The IMP/TIP software was a pain to re-write: we did it once for the Pluribus and that was enough]. In the small computer world this may mean more machines with 6502's grafted onto the same bus as 8088s and 68000s. What kinds of problems does this cause? Well, it's a thought. Rick. [smith.umn-cs@CSNet-Relay] [...ihnp4!umn-cs!smith]
henry@utzoo.UUCP (Henry Spencer) (04/27/84)
People have built systems with multiple processors in the same box,
sharing some resources. The processors generally have their own
memory over and above any shared memory, because memory contention
can degrade processor performance pretty quickly. There is some
economy of scale in communications (backplanes are -- or at least,
look -- cheaper than local networks) and in support services like
power supplies and mechanical hardware. My impression, though, is
that the savings are not huge, and that the people doing it are
mostly folks who wouldn't know good local networking if it bit them.
--
Henry Spencer @ U of Toronto Zoology
{allegra,ihnp4,linus,decvax}!utzoo!henrysmith@umn-cs.UUCP (Richard Smith) (04/30/84)
#R:houxk:-25300:umn-cs:13900010:000:697 umn-cs!smith Apr 28 22:32:00 1984 Re: Multiprocessors Most of what Henry says about multiprocessors is true, but I don't think that using backplanes is a good alternative to local nets. I tend to think of a 'backplane' as representing a parallel communications path that ALSO distributes power; managing the power supplies sounds like a source of real trouble. Imagine trying to provide clean power (incrementally!) in a system as it grows from say 10 to 1000 nodes... The thing I like about local net architecture is that we can separate everything into individually cheaper and isolated boxes. You do lose the economies of scale with the power supplies, though. Rick. [smith.umn-cs@CSNet-Relay] [...ihnp4!umn-cs!smith]
wmartin@brl-vgr.ARPA (Will Martin ) (05/01/84)
I have a different topic, but one that's probably a bit simplistic for the level of this readership. A recent message mentioned "VAX-class superminis" (or similar wording). Our organization is soon to go out on procurement for a VAX-type machine, 780 or equivalent. I really don't know what is out there that is "equivalent" to a 780 (that means same or better performance for the same or less money). We know the 780 is being superseded by the 785, 790, etc; we also know that there are other manufacturers' machines, like the 3B20 or CCI's "Tahoe". (Is the Tahoe the same as the Power 6/32 mentioned elsewhere?) What else is out there that might show up in response to such a procurement solicitation? It has to run UNIX (either 4.2 BSD or System V), and will be used for office automation user support, not number-crunching. I've heard of the Ridge 32, but know nothing about it but the name. If any commercial establishments on the net want to send me literature on their products, that's fine with me -- my US Mail address is: W. G. Martin USArmy DARCOM ALMSA ATTN: DRXAL-RI PO Box 1578 St. Louis, MO 63188 Regards, Will
guy@rlgvax.UUCP (Guy Harris) (05/02/84)
> We know the 780 is being superseded by the 785, 790, etc; we > also know that there are other manufacturers' machines, like > the 3B20 or CCI's "Tahoe". (Is the Tahoe the same as the Power > 6/32 mentioned elsewhere?) Yes. We like to use lakes as code names for our processors; our small 68K box was named after a small, polluted lake (Oneida) and a multi-processor 68K system was named after a large, polluted lake (Erie). We are now tossing around the idea of a smaller system and have been using the name "Walden"... :-) Guy Harris {seismo,ihnp4,allegra}!rlgvax!guy