I'm trying to track down a possible bug in my Linux-AFS
implementation. The symptom that people are seeing is that they try
to talk to a server which really is up but AFS reports it as lost
contact. Next time the servercheck is done, it comes up and is fine.
(In the process, they sometimes get an error 110 while trying to store
a file -- error 110 is VBUSY).
I was wondering if AFS depends upon any specific behavior of the
Kernel Network code. In particular, does it depend on any special
behavior of the network write function (i.e., blocking, no dropped
packets, etc).
Linux, it seems, tends to drop packets on the floor alot (at least
from what I'm told), and when it does this, AFS gets very unhappy.
Anyone have any insight for me?
Thanks.
-derek
