home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!contex!marvin.contex.com!frank
- From: frank@marvin.contex.com (Frank Perdicaro)
- Newsgroups: comp.sys.sgi
- Subject: Re: vhand is bringing my R4K Indigo to its knees!
- Message-ID: <3272@contex.contex.com>
- Date: 12 Jan 93 13:41:21 GMT
- References: <1993Jan9.003815.18504@ringer.cs.utsa.edu> <ul2uirg@zuni.esd.sgi.com>
- Sender: news@contex.contex.com
- Organization: Xyvision Design Systems
- Lines: 36
-
- In article <ul2uirg@zuni.esd.sgi.com> olson@anchor.esd.sgi.com (Dave Olson) writes:
- >In <1993Jan9.003815.18504@ringer.cs.utsa.edu> senseman@ricky.brainlab.utsa.edu (David M. Senseman) writes:
- >| Our newly upgraded R4K Indigo is SLOWER than our 4D/50 GT :-P
- >Perhaps you are trying to run just a few more/larger apps ;)
- >| Now what did I screw up?
- >Well, it looks like you are running far more jobs than you have
- >memory for, so you are doing a lot of paging.
- >Something is *seriously* wrong with your ksh processes. How could
- >you possibly have ksh's with 3000 pages (12 MB) resident ?!?!?
-
- Before SGI gave away ksh, we used an AT&T ksh port. There was a bug
- in ksh due to a bad cast someplace that caused similar problems.
- Something about a terminal read getting confused under particular
- conditions, and instead of getting -1, it got 2^32, and tried to
- execute some function chain 2^32 times. On occasion, if the system
- load was correct ( code not localized, high system demand --> swapping,
- paging ), this would virtually halt a system. Ocassionally, I see
- piggish ksh behavior today, and wonder.
-
- It does not look like your problem.
-
- An example of the effect memory can have. We had some of the origanal
- 8k/8k 4D20s, one with 8mb. A process was stared on it, and after a while,
- we estimated it would take 40 hours, based on how much was done.
- The same code was started on a 4D25 with 16mb and the 32k/64?k. The
- job was finished in less than 10 minutes. The effect of extra memory
- can be highly non-linear, and highly non-intuitive, and is therefore
- often discounted by folks that have not seen wild changed in program
- behavior based on memory.
-
- Get some more memory.
- --
- Frank Evan Perdicaro Xyvision Color Systems
- Legalize guns, drugs and cash...today. 101 Edgewater Drive
- inhouse: frank@marvin, x5572 Wakefield MA
- outhouse: frank@contex.com, 617-245-4100x5572 018801285
-