home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.sys.sgi
- Path: sparky!uunet!cs.utexas.edu!sun-barr!ames!sgi!fido!zola!sgihub!sydney.sgi.com!news
- From: mitch@rama.sydney.sgi.com ()
- Subject: Re: is 4D/240 VGX w/ IRIX 3.3.2 SMP
- Message-ID: <1992Dec17.021049.28137@sydney.sgi.com>
- Sender: news@sydney.sgi.com (System News Owner)
- Organization: Silicon Graphics Computer Systems, Mtn. View CA. 94039
- References: <Bz0n49.1yHM@austin.ibm.com> <1g6a2lINN6mv@fangio.asd.sgi.com>
- Date: Thu, 17 Dec 92 02:10:49 GMT
- Lines: 51
-
- In article <1g6a2lINN6mv@fangio.asd.sgi.com>, rck@fangio.asd.sgi.com
- (Robert Keller) writes:
- |> dimpsey@austin.ibm.com writes:
- |> >I have just received access to a SGI IRIS 4D/240 VGX running IRIX
- |> >system V REL. 3.3.2. Is this a symmetric multiprocessor or are the
- |> >processors following some master-slave configuration.
- |>
- |> All SGI multiprocessor machines are SMP.
- |> >When I log in am I always on the same processor?
- |> No. Furthermore, your shell process (any process, really) can
- |> seemlessly switch processors at any time.
- |>
- |> >Can I bind a process to a processor??
- |>
- |> Yes, checkout the mpadmin(1) man page.
- |>
- |> ...robert
-
- Please read the mpadmin page with care.
-
- To some degree there are no good reasons to mpadmin
- anything.
-
- The mpadmin and other hooks can make the machine look hung.
- Users who mpadmin and npri processes are taking exact control
- of a complex system. More problems are created than solved
- in my experience. Since the trouble options are uid 0 restricted
- there are almost never problems unless a system manager is careless.
-
- They are very useful in some realtime contexts. On systems with
- four or more processors it is often useful to tinker with restricting
- the last couple of processors while leaving 0 and 1 to free wheel.
-
- In terms of process scheduling there is one runq each processor
- obtains a reliable hardware lock. Finds work to do on the runq
- pulls the process off the Q frees the lock and runs the process
- untill the next time slice clock tick.
-
- Then the runq is locked the last job placed back on the q. The
- q is then inspected and the 'next' is picked off to run. This part
- is very vanila SysV inside the lock.
-
- Below this are more complex MP issues, but from a users point of view
- there is only 1 runq. Look at "ps -elf" listings. There is a P
- field to indicate which processor a process happens to be on.
-
-
- Enjoy,
- mitch
-
-
-