home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!mcsun!sun4nl!tuegate.tue.nl!svin09!wzv!wietse
- From: wietse@wzv.win.tue.nl (Wietse Venema)
- Newsgroups: comp.sys.sun.admin
- Subject: Re: Sun 690MP performance problems
- Message-ID: <3843@wzv.win.tue.nl>
- Date: 2 Sep 92 21:25:04 GMT
- References: <2A9FA7FE.2305@orion.oac.uci.edu> <2436@tymix.Tymnet.COM>
- Organization: Eindhoven University of Technology, The Netherlands
- Lines: 21
-
- feldman@comet.Tymnet.COM (Steve Feldman) writes:
-
- >>in "D" state when the system performance gets bad. Response time for
- >>online users gets slow - people using the Gnu Emacs mhe mail package
- >>wait a long time (minutes in some cases) for response to commands, "df"
- >>takes up to a minute to complete, etc.
-
- >I have seen problems similar to this on a Sparcstation 1+ which is our
- >campus mail gateway and news server. This machine gets very little
- >interactive use, and the mail load is fairly light, so most of what it does
- >is news processing and nntp. It has two SCSI disks. The symptoms sound
- >remarkably similar to what you've reported, but since our machine isn't
- >used as a timesharing system I haven't worried about it too much.
-
- In this case, you are probably running up against the max number of
- contexts that the hardware can handle (8-16 for the smaller sparcs, 64
- or more for other models). When the number of active processes exceeds
- this limit, context switches become extremely expensive and the machine
- crawls to a halt.
-
- Wietse
-