home *** CD-ROM | disk | FTP | other *** search
- Xref: sparky comp.arch:9402 comp.parallel:2119
- Newsgroups: comp.arch,comp.parallel
- Path: sparky!uunet!gatech!hubcap!fpst
- From: barmar@think.com (Barry Margolin)
- Subject: Re: sgi/sun shared memory machines question
- Message-ID: <18qtpoINN57g@early-bird.think.com>
- Sender: fpst@hubcap.clemson.edu (Steve Stevenson)
- Nntp-Posting-Host: telecaster.think.com
- Organization: Thinking Machines Corporation, Cambridge MA, USA
- References: <1992Sep11.121830.17069@hubcap.clemson.edu>
- Distribution: usa
- Date: 11 Sep 1992 19:58:48 GMT
- Approved: parallel@hubcap.clemson.edu
- Lines: 30
-
- In article <1992Sep11.121830.17069@hubcap.clemson.edu> zilla@nj.nec.com (John Lewis) writes:
- >Someone told me that the SGI multiprocessor machines cannot run
- >different unix processes on different processors; they can only
- >run a single (but possibly parallelized) process at a given time.
- >The reason was said to be that the machines do not implement
- >any mechanism for simultaneous writes.
- >
- >Is this correct? If so, how are simultaneous writes from
- >different threads of a parallelized process prevented?
-
- I don't know whether it's true or not. But if it is, prevention of
- simultaneous writes within a process can be prevented using a semaphore in
- shared memory. The write() routine would check this, and delay until the
- semaphore is cleared.
-
- >Is the new sun multiprocessor (4/690?) similar?
-
- No, it's exactly opposite. A process can only be on one processor at a
- time, so each processor runs a different process.
-
- Prevention of simultaneous writes is accomplished by only allowing one
- processor to be in the kernel at a time. Upcoming releases of the OS
- removes this restriction, and uses locks around specific critical regions
- of the kernel rather than the entire kernel.
- --
- Barry Margolin
- System Manager, Thinking Machines Corp.
-
- barmar@think.com {uunet,harvard}!think!barmar
-
-