it seems that nfs over tcp is not reliable if the client is a 2.1.103 alpha
box. The server is a Solaris 2.5.1 Ultra1/140 (yeah, I know, it should be
under ultrapenguin, but I am not responsible for this one). When I write
a 4 MB file, only 57344 bytes make their way to the server, and the writing
process goes into an annoying D state, so I cannot kill it and therefore
cannot remount the partition using the more functionnal udp.
By the way, I still have performance problems when writing on a udp-nfs
mounted partition (same server, same client) : I read at 1.1 MB/s, and I
write at 200 KB/s. A Solaris client, on the same ethernet and with the
same server, read and writes at 850 KB/s. I am willing to help correcting
this problem (for I want to convince my sysadmin that nfs is good on linux,
and therefore at least old SunOS 4.x stations should be converted to
ultrapenguin). I have root access on the client (not the server) and I
can reboot it rather often for the moment.
I tried the "noac" mount option, it does not change anything.
Thanks for any help.
--Thomas Pornin
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu