ghfeil@white.toronto.edu (Georg Feil) (12/29/89)
We are at wit's end trying to solve problems of long delays and timeouts
when remote NFS servers or the intervening networks go down.
In particular, we mount /usr/local from a remote server that tends to go
down frequently. Since /usr/local/bin appears fairly early in the search
path, users suffer long delays for each command typed while NFS retries
the downed server and eventually times out. News is also mounted remotely.
A related problem is that is takes extremely long to log on in this
situation. Progress of the login halts just after printing /etc/motd.
Our system consists of a Sun 3/50 and Sun 3/280 server running OS 4.0.1.
Our local Ethernet is connected to a remote network using a line-of-sight
laser link. That's right. It snows, it rains, it fogs, and the link goes
down. We have experimented with mount options in the /etc/fstab file
until arriving at the following:
apl:/ists /usr/local nfs ro,soft,intr,bg,retry=15,timeo=5,retrans=3 0 0
By my calculation, timeouts should take about 7 seconds total with this
configuration. Nonetheless, delays range from 1 to several minutes. Then
the following message typically appears:
NFS getattr failed for server apl: RPC: timed out
We have tried making /usr/local a symbolic link to the actual mount point.
Someone suggested two levels of symbolic links, but this made no
difference either. Any suggestions or hints anyone can provide would be
appreciated.
Thanks,
Georg.