On Sat, 2 Jan 2010, Jeremie Le Hen wrote:
I will reduce the MTU and see if the problem arises with directories that could be read correctly otherwise. I will keep you informed.
You can reduce rsize to 1024, so that it isn't happening, but that will also change the size of the readdir reply. (ie. It might make the problem go away, but doesn't prove that IP fragmentation is the cause..)
Does NFS permit to fragment/reassemble at the application layer?
Nope. IP fragment reassembly should work fine and IP fragmentation happens all the time when using NFS over UDP. (Some would say NFS over UDP misused/abused IP fragmentation. In any case, it happens more frequently there than anywhere else.)
ps: I vaguely recall that the problem went away when you used TCP instead. Is that correct?Yes this is correct.
Thanks for poking at this, rick _______________________________________________ [email protected] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-stable To unsubscribe, send any mail to "[email protected]"
