On Thu, 2014-08-07 at 12:46 -0500, Andrew Deason wrote:
> On Wed, 06 Aug 2014 15:33:02 -0400
> Dale Pontius <[email protected]> wrote:
> 
> > Obviously this was client side, but I find it hard to believe that
> > keeping a connection mapped for the 2 hours mentioned elsewhere would
> > be necessary.
> 
> Maybe not "necessary", but at least in the past it was possible for
> fileserver -> client communication to occur several hours after the last
> client -> fileserver communication. Jaap Winius found this by
> experimenting with the port mapping timeouts on his equipment:
> <https://lists.openafs.org/pipermail/openafs-info/2011-May/036014.html>

More specifically, cases where only one client is touching a volume or
that volume is read only won't cause problems; nor will many
modifications coming often from multiple clients, since normal NAT
timeouts will be sufficient except in rare cases (like the time I recall
from years ago when someone tried using AFS behind a commodity NAT
router with something like a 10 second UDP NAT timeout. That one was a
double whammy because of both the short timeout and that NTP had fallen
into a pattern that was just outside the timeout, causing the small UDP
NAT table to overflow in about half an hour, if I recall correctly).
It's when they are occasional that you can run into problems --- and
these problems will often not be obvious, since they will only have a
noticeable effect sporadically and only some time after the actual
triggering event, which is pretty much the worst possible case for
detecting and recognizing them.

-- 
brandon s allbery kf8nh                           sine nomine associates
[email protected]                              [email protected]
unix openafs kerberos infrastructure xmonad        http://sinenomine.net

Reply via email to