svn commit: r197298 - head/sbin/mount_nfs

[Rick Macklem]

On Sun, 20 Sep 2009, M. Warner Losh wrote:

> : Someone mentioned an issue w.r.t. umount using UDP, which was basically
> : a slow timeout when a server didn't handle the UDP call, such as when
> : it was down. I'd guess that a TCP attempt would fail more quickly than
> : a UDP attempt when the server doesn't have rpcbind/portmapper running.
> : (Not so sure when the server has crashed, but I'm guessing that the TCP
> : connection attempt fails more quickly than the N retries over UDP?)
>
> A server down would have the same timeout issues as a UDP timeout.
>
I was thinking of this email:
http://docs.freebsd.org/cgi/getmsg.cgi?fetch=213665+0+archive/2009/freebsd-current/2

Actually, we were both wrong (although you were closer), at least for the
little experiment I just tried.:-)

When the server is up, but not supporting that protocol (UDP or TCP), both
cases failed in < 1sec. (So, you got this one right. I had thought UDP
would take longer, but it didn't. It appears that the ICMP Dest 
Unreachable causes it to fail quickly. I have no idea if ICMP Dest.
Unreachable messages get forwarded normally. Maybe that's what the
email was alluding to when he said "some situations".)

When the server is down/network partitioned, it actually took longer
for TCP (1min vs about 10sec) for the defaults in the 8.0 RPC libraries.

Probably not relevent to the discussion, but I thought someone might
find it interesting, rick