home *** CD-ROM | disk | FTP | other *** search
- Xref: sparky comp.sys.atari.st:12430 comp.sys.atari.st.tech:4452
- Path: sparky!uunet!usc!chaph.usc.edu!news
- From: baffoni@aludra.usc.edu (Juxtaposer)
- Newsgroups: comp.sys.atari.st,comp.sys.atari.st.tech
- Subject: Re: Utterly bizarre idea for Atari
- Date: 17 Aug 1992 13:28:57 -0700
- Organization: University of Southern California, Los Angeles, CA
- Lines: 49
- Message-ID: <l90309INN6q5@aludra.usc.edu>
- References: <1992Aug15.043618.17054@news.csuohio.edu>
- NNTP-Posting-Host: aludra.usc.edu
-
- In article <1992Aug15.043618.17054@news.csuohio.edu> max@madnick.cba.csuohio.edu (Max Polk) writes:
- >This is an utterly bizarre idea that will either send you into a
- >screaming fit or drop your diastolic blood pressure ten or twenty
- [idea of making a 68k based multiprocessor Atari machine removed]
-
- This sounds like a good idea, except:
-
- 1) The 68ks were not built for multiprocessing (big surprise) - you are not
- only talking about totally (almost) re-writing TOS to handle the multiple
- processors (which sounds like a good idea to me anyway), but you would have
- to implement some impressive glue-logic to get the multiple processors to
- work cooperatively since they a) don't know that they don't own the bus, b)
- don't understand any kind of memory but super/user, c) only have 24bits
- address line (and you really want to be able to share some of the memory so
- you can at least pass information between processes/virtual machines).
-
- 2) Related to the last point of 1), you really want to have a distributed
- multiprocessor-system - how many times do you run 27 applications simul-
- taneously? Not to mention that each processor would have to have its own
- memory - and it doesn't sound too appealing or marketable to have to buy
- 27MB so you can have 27 1-MB virtual machines. That and the fact that each
- would still be going at the original (single processor) speed. By having
- a distributed multiprocessor system, you distribute tasks over the multiple
- processors so that you can have (with properly written software) a (several)
- programs executing in parallel over as many processors as is possible/feasable.
- Therefore, you get a faster execution time with a slower (=cheaper) technology.
-
- My suggestions:
-
- To address 1) above, use a 68040 as a minimum - you need not get a faster one,
- (<=25MHz should be more than enough, and with the '040/33, prices should be
- dropping) - this would be an excellent choice of multiprocessor as it has
- built in cache w/cache snoop (if something on the bus requests data that is
- cached on the '040, it supplies the data), MMU so that memory can be
- partitioned however needed (ie. processor N's memory, sharable, etc.), and
- it can address 32bits of dataspace so that a huge chunk of memory can be
- partitioned dynamically to the processors. Most importantly, the '040 does
- not bus-master the bus - it requests access to the bus just like any other
- device on the bus, thus making it nice for other processors that might be
- present.
-
- So how does that sound? Atari - feel like making some low-cost
- multi-processor workstations?
-
-
- -Mike
-
-
-
-