This may be a result of changing nfs code in the kernel, but
it would be nice to know what the problem is and when it is
likely to be fixed.
Basically, copying a file from an nfs mounted volume to a
local disk (both client and server are RH6 machines) is more
than 3 times slower than ftping the same file. Sometime the
nfs transfer just stops for "long" periods of time. The
machines I'm working with are on the same network segment on
an unloaded network.
I've already tried using the rsize and wsize mount options
for nfs. They have little to no effect.
Forgot to make this high priority/severity. This is really critical
as I am going to have to stop using my Linux workstation and go back
to Windows and smb to get acceptible peformance from our Linux file
Linux-smb-Linux does not work for mounting home directories.
Permissions, file name case, and dot files are all messed up. You
have to use nfs not smb for home directories.
Are you running knfsd? We do not see this kind of performance
degradation here in the office.
------- Email Received From Bill Chatfield <bill.chatfield> 06/29/99 11:09 -------
I do believe we are running knfsd because lsmod says I have an nfs
kernel module loaded.
I would be happy to provide telnet access to a Red Hat person so you
could see the problem where it is happening. Just let me know.
BTW: Going back to using Windows was a threat I couldn't live up to.
The NFS is slow, but it should not be that painfully slow. What is the
size of the file you are dealing with?
On a related thing, Samba can be instrumented to preserve the case for
filenames, it just does not do it by default.
On a 100mbit network the NFS server will peak at about 1.9MB/sec,
while ftp is able to go up to 5 or 6 MB/sec, so dependong on the type
of transfers that you are doing you will see the type of slowdown that
you are talking about.
I originally thought the machines in question were on the same physical subnet,
but after physically tracing the ethernet wires, it appears that they are not.
I suspect that this is the cause of the slowness. Does that make sense?