After experimenting with the 2.1.88 nfs client some more, it seems that
the dentries are being flushed after a change in mtime of the directory on
the server. To see this behaviour, run vmstat on one console and then cat
dir/bigfile >/dev/null on another. The file will remain in the cache
until, say, ls -d dir is done, or even another cat dir/bigfile. As an
alternative to blasting each directory entry, shouldn't the client mark
each entry as needing validation? The server is "Universal NFS Server
2.2beta16" running on 2.0.31+patches, ext2 filesystems with atime enabled
running on a p133/32megs, client being a dual p200mmx/32megs (almost all
of which is free).
Also, try doing an rm -rf `ls | sort -r` on the server while an rm -rf on
the same directory on the client has already been running a while - the
client can end up in an infinite loop spamming the console with a message
I can't quite recall at the moment.
-ben
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu