kwongj@caldwr.caldwr.gov (James Kwong) (01/14/88)
NFS Problem
S.W. ver- Aegis 9.7, BSD4.2 9.5, TCP/IP 3.0/w patches, NFS 1.0
Machines: DN3000s, SUN 3/160
Apollo's implementation of NFS has some outstanding features such
as allowing for the "//" directory to be mounted on the foreign
host, making all the nodes available to it with one mount point;
mounting the foreign host on one of the Apollo root directory,
and with the ns_helper running, allows the other Apollo nodes
to 'cd' or 'wd' to the foreign host with one set of NFS server
running.
But...
From our experience, Apollo's NFS Ver. 1.0 has a history of un-reliable
services. The NFS daemon (nfsd) stops functioning after a several hours;
some time after several days. When it 'dies' it sometimes takes the
tcp_server with it (or is it tcp_server thats dies and drags nfsd
down ?)
Examination of the `node_data/nfs_error_log shows bunch of:
'.
.
.
********************************************************************************
Entry created on 12-Jan-1988 13:00 Node id: E298 Process name: nfsd-1.01
NFSD: cvrt_status()..FYI..Unable to convert aegis error number e0002H. - directory is full (OS/naming server)
.
.
.'
which baffles us, since disk space free on the Apollos node
are from 25 to 50 percent and the Sun is around 5 percent.
Also when rebooting the node, the nfsd sometimes starts up functionally.
The other times, it starts, but it is unresponsive to the remote foreign
node's polling of the nfs server in which case we have to kill the
process on the Apollo end and restart it manually.
Several questions:
The client hosts need not run any nfs server (i.e. portmap or
nfsd). Just install nfs on the host as a client install. Correct ?
----------------------------
The correct? way to start up the nfs server is in the node startup
file (e.g. /sys/node_data/startup.1280bw) with dm commands:
'cps /etc/portmap -s10'
'cps /etc/mountd -s15'
'cps /etc/nfsd -s15'
(The -sn option puts the nfs srver to sleep for n seconds
to allow for tcp/ip server to fully initialize.)
and NOT:
'cps /etc/portmap -n portmap'
and so on.
(release note says former; instruction after installation says
latter.)
Or it doesn't matter which way it's started?
---------------------------
Can the remote host /etc/passwd file be a subset of the Apollo
/etc/passwd file. Both having the same uid for the users in
common. Ditto for the /etc/group file.
---------------------------
Does heavy network traffic takes its toll on the NFS dameon?
---------------------------
All is 'normal' for this version; just sit tight and wait for the next release?
James Kwong
Sacramento, CA 95802
Calif. Depart. of Water Resources ucdavis.edu!caldwr!kwongj (Internet)
...!ucbvax!ucdavis!caldwr!kwongj (UUCP)
"Our program who art in memory, HELLO be thy name.. "
Disclaimer:
The opinions expressed above are mine, not those of the State
of California or the California Department of Water Resources.