--On Sunday, April 27, 2008 09:45:34 AM -0400 Jeffrey Altman
<[EMAIL PROTECTED]> wrote:
Simon Wilkinson wrote:
Tune the AFS timeouts so that the user get's failure messages more
quickly?
If by "tune the AFS timeouts" you mean "make them smaller", don't do that.
It is tempting to lower timeouts to reduce the amount of time you have to
wait when something isn't working, assuming that when it is working you'll
get a response quickly because it seems that way every time you look. But
most people haven't seen a wide enough variety of networks to make this
generalization; a "short" timeout that works for you may not work so well
for someone with a high-latency path or a heavily-loaded server, and
setting timeouts a bit too short means more and more frequent
retransmissions, which can make a bad situation (heavy congestion) worse.
Performing the check at the time the network configuration changed will
ensure that the cache manager knows as soon as possible whether or not
the desired servers are in fact accessible. If they are not the cache
manager can fail requests immediately.
Performing the check at the time the network configuration changes will
guarantee that the cache manager talks to a large number of servers
(normally, every server it has ever had a reference to since starting) many
of which the user may not care about.
It seems the simplest thing to do here is whenever an interface is brought
up to reset the state of all servers to "up". This results in a fast
response time for a server that really is up, at the expense of an extra
timeout in cases where the user touches a server that is down before the CM
gets around to doing the next 5 minute check.
-- Jeff
_______________________________________________
OpenAFS-devel mailing list
[email protected]
https://lists.openafs.org/mailman/listinfo/openafs-devel
