home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.sys.sun.admin
- Path: sparky!uunet!mnemosyne.cs.du.edu!aburt
- From: aburt@mnemosyne.cs.du.edu (Andrew Burt)
- Subject: Re: Amd is not perfect
- Message-ID: <1992Aug14.163858.8018@mnemosyne.cs.du.edu>
- Keywords: nfsd automount amd D
- Organization: University of Denver, Dept. of Math & Comp. Sci.
- References: <Y7BF99A9_@linac.fnal.gov> <1992Aug5.231434.20156@trl.oz.au> <13916@auspex-gw.auspex.com> <1992Aug9.184414.4765@newshost.lanl.gov>
- Date: Fri, 14 Aug 92 16:38:58 GMT
- Lines: 56
-
- In <1992Aug9.184414.4765@newshost.lanl.gov> dlc@c3serve.c3.lanl.gov (Dale Carstensen) writes:
- >I think most of the responses to the original posting in this thread
- >have been about 2 separate problems:
-
- > 1. nfsd's go into D status (a server-end problem)
-
- > 2. automount (or possibly amd, and amd probably has a different
- > problem than automount since amd forks child processes to
- > handle new mounts) blocks, so any process requiring a new mount
- > will go into D status waiting for automount (a client-end
- > problem)
-
- >...one thing I haven't seen a load factor go beyond maybe 6 or 8 due to
- >these 2 problems, and other messages have mentioned 400+. But maybe
- >problem 2 could generate quite a lot of load -- my discussion below
- >involves a rather quiet Saturday morning as the peak load.
-
- I've seen this (often) in our config:
-
- Pyramid 90x with OSx4.0 (aka "ancient") acting as disk server to
- Sparcserver 2 with 4.1.2
-
- The Pyramid will reboot (for other unrelated causes) and the
- the Sparc's nfsd will never see it come back up(?) -- instead, they
- get stuck in "D" and the load shoots waaaaay up (I've seen a load of >300
- if the Pyramid re-crashes or stays down; eventually the sparc dies too).
-
- We don't run any automounter. Granted the Pyramid's NFS is old, but
- the SunOS is new.
-
- Note, that if the Pyramid reboots, sometimes the sparc STILL doesn't
- notice that it's up, BUT rebooting the P. wakes it right up.
-
- >... There also was a bug in 4.1 that would cause all
- >nfsd processes to go into D status when an exported filesystem
- >filled up, which was fixed by the NFS jumbo patch for 4.1, I think.
- ^^^^^^^^^
-
- Which is frequent at this site -- but note the sparc runs 4.1.2.
-
- >Problem 2 I've had daily for 2 days, and now that I understand it,
- >here's what is was for me this time. My NIS master server has been
- >hanging processes in D status, especially rlogin daemons.
-
- Same here re rlogin's -- to the point that all ptys get used up
- until the NFS problem is resolved.
-
- My solution is to modify the login procedure so it checks to see if the
- home-dir server is up via ping. (Easy since users log into a shell script.)
-
- However, the real problem remains unsolved -- fixes still appreciated...
- --
-
- Andrew Burt aburt@du.edu
-
- "And that, my liege, is how we know the Earth to be banana-shaped"
-