home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!pipex!warwick!str-ccsun!strath-cs!baird!jim
- From: jim@cs.strath.ac.uk (Jim Reid)
- Newsgroups: comp.protocols.nfs
- Subject: Re: NFS errors; "killed on swap error"; "illegal instruction"
- Message-ID: <JIM.92Sep4120307@hunter.cs.strath.ac.uk>
- Date: 4 Sep 92 11:03:07 GMT
- References: <1992Sep3.015551.26268@usage.csd.unsw.OZ.AU>
- <1992Sep3.210948.28944@decuac.dec.com>
- Sender: news@cs.strath.ac.uk
- Organization: Computer Science Dept., Strathclyde Univ., Glasgow, Scotland.
- Lines: 21
- Nntp-Posting-Host: hunter
- In-reply-to: mjr@hussar.dco.dec.com's message of 3 Sep 92 21:09:48 GMT
-
- In article <1992Sep3.210948.28944@decuac.dec.com> mjr@hussar.dco.dec.com (Marcus J. Ranum) writes:
-
- >Does anyone else have lots of read/write errors with NFS? I get several
- >per day. I suspect that the problem is that the server (DECstation
- >5000/200, ULTRIX 4.2A) is very heavily loaded (~50 users typically),
- >although I would have thought that this would have resulted in timeouts
- >(which I also see) rather than read/write errors. Any comments?
-
- Do you have the filesystems mounted "soft"? If the server is
- heavily loaded and an NFS request times out it'll give a write error
- on "soft" mounts, rather than retrying. If that error happens to be
- a page of an executable that's being paged from the text image, you
- will get a "swap error" and it'll kill the process.
-
- I understand that some NFS implementations have/had a nasty bug with
- soft-mounted file systems. Instead of returning an error when a read
- request timed out, the code would sometimes pretend that it got a
- valid block of data - all zeros! This is very probably not what you
- want to happen when paging in an executable over NFS....
-
- Jim
-