home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.sys.sgi
- Path: sparky!uunet!cs.utexas.edu!sun-barr!ames!sgi!rhyolite!vjs
- From: vjs@rhyolite.wpd.sgi.com (Vernon Schryver)
- Subject: Re: NFS Question...
- Message-ID: <p97tva0@rhyolite.wpd.sgi.com>
- Organization: Silicon Graphics, Inc. Mountain View, CA
- References: <1992Aug31.165654.24006@news.iastate.edu>
- Date: Mon, 31 Aug 1992 19:54:37 GMT
- Lines: 55
-
- In article <1992Aug31.165654.24006@news.iastate.edu>, bryan@pi.eai.iastate.edu (Bryan Manske) writes:
- >
- > If a line in my fstab looks like:
- >
- > [machinename]:/usr/mp1 /usr/mp1 nfs bg,soft,timeo=28,retrans=16 0 0
- > ^^^^^^^
- >
- > should the 'bg' and 'soft' options prevent network lockups when one
- > machine (nfs server) goes down? We have several sgi machines cross
- > mounted and it seems that whenever one goes down the hard way (and
- > sometimes even when its brought down nicely) the rest of the machines
- > lock up. That is to say, you can log in, but get no prompt in any
- > of the shells.
- >
- > Man pages lead me to believe that I might need different values for
- > timeout and retransmission.
-
-
- Read `man fstab`.
-
- You have told your client to start with a time-out of 2.8 seconds, and
- exponentially back off 16 times. That is a not much different from
- a hard mount.
-
- Almost everyone who changes timeo and retrans gets them wrong.
- Almost everyone who changes them would be better served by the
- default values.
-
-
- The problem you are having with not getting prompts from shells
- is because you are mounting in /usr. Many commands in the system,
- and especially shells, do a lot of the equivalent of the /bin/pwd.
- That involves the algorithm:
- 0. start with the directories x=.. and y=.
- 1. lstat(2) each file in x, looking for the major, minor and
- inumber of y, when found, go to step 3.
- 2. note the name corresponding to the file in x found in step 1,
- adding it to the growing path.
- 3. if the name saved in 2 was "/", exit
- 4. set x=the new .. and y=the old x
- 5. go to step 1
-
- Notice that step #2 involves stat'ing every file in a directory until
- the target is found. If one of those files is a dead NFS mount,
- the algorithm will stick, and if you are using a hard mount or
- your time-out parameters, it will stick approximately forever.
-
- The solution is either to use the automounter and replace /usr/mp1
- with a symbolic link /usr/mp1->/hosts/machinename/usr/mp1,
-
- or to put all of your symbolic links into some other directory by changing
- your fstab, say /net, and replace /usr/mp1 with a symbolic link to /net.
-
-
- Vernon Schryver, vjs@sgi.com
-