home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!mcsun!uknet!ox-prg!michael
- From: michael@uk.ac.oxford.robots (& Stevens)
- Newsgroups: comp.sys.transputer
- Subject: Re: Mixed C40/transputer networks & 3L Parallel C
- Message-ID: <MICHAEL.92Aug19213930@lucrece.uk.ac.oxford>
- Date: 19 Aug 92 20:39:30 GMT
- References: <2a9282d0@ThreeL.co.uk>
- Sender: news@comlab.ox.ac.uk
- Organization: Dept. Engineering Science, Oxford University, UK
- Lines: 31
- In-reply-to: adc@threel.co.uk's message of 19 Aug 92 16:20:16 GMT
-
- Just thought I would add my two pence worth to debate.
-
- Since 3L Parallel(ha ha) C never did support fine grained
- parallelism I guess it doesn't matter too much what you run it on.
- Transputer, C40, even 680x0 using semaphores and shared memory: it
- would all be much the same. At least Inmos just call a C with a
- process library "ANSI C" and make no claims for its parallel nature.
-
- Never having used a C40 (this being comp.sys.transputer) the
- following questions come to mind.
-
- 1. Its a newer chip then a T8 and clearly faster, but it an actual
- multiprocessing system do you get more bang per buck?
-
- 2. Is the architecture up to the job? Context switch time is just as
- important as comunications speed. To get the much sort after linear
- increase in performace
-
- Performance | /
- | /
- | /
- |/
- |--------
- No of processors
- each processor must be able to fill up all its available time. To do
- this it must be able to quickly do something else while it waits for
- other processors i.e. Context switch
-
- --
- Michael Stevens, Robotics Research Group,Dept of Engineering Science,Oxford,UK
- INTERNET: michael@robots.oxford.ac.uk
-