On tuesday I asked about how to NFS mount a file system from a Convex
file server, which migrates files to a tape robot if not accessed recently.
The problem was not the mount itself or the access to files on the disk,
but access to migrated files, because they can take 10 minutes to get loaded.
Many thanks to
mike@birch.lib.utexas.edu (Michael Kline)
Daniel Pommert <daniel@lternet.edu>
ye@Software.ORG (Jian Ye)
poffen@San-Jose.ate.slb.com (Russ Poffenberger)
Ray Schnitzler <ras@unipress.com>
Kerien Fitzpatrick <fitz@frc2.frc.ri.cmu.edu>
jdschn@nicsn1.monsanto.com (John D Schneider)
mbecker@elf.igraph.com (Mark Becker)
grefen@connie.de.convex.com (Stefan Grefen)
emiller@convex.com (Evan Miller)
The most detailed answer came from within Convex, and is the following :
NFS has no idea about migration. For NFS this is like a (very) slow server.
> The complete options I use are timeo=3000,rw,hard,intr,noquota,nosuid.
I think thats part of your problem. You block each biod up to 300 seconds or
until the file ist staged. With read-ahead this can block all biod's reading a
single file.
Increase the number of biod's from 4 to 16 and try timeo=5,retrans=600.
This may help you. (It does help on other systems but each NFS implementaion
is different).
> Do I have to use "soft" ? I thought only "hard" would ensure correct
> error handling in case of write errors for rw nfs file systems ...
Use hard mounts, soft mounts would fail completly.
I tried this out and it seems to have solved my problems.
Eckhard R|ggeberg
eckhard@ts.go.dlr.de
This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:08:58 CDT