Dave,

Thanks for your detailed read and comments.  Going to try to
incorporate them into the next update.

2009/7/14 Craig, Dave <[email protected]>:
> 9. In 5.6.2.2.1. fifth paragraph, was the following intended?  "If all 
> retransmissions for a message fail, _the sending node should close the 
> connection routing the message_."  Should there also be additional hysteresis 
> to make sure that a failing connection is not quickly reestablished under a 
> fast chord-reload-ping-frequency?
>

Yes, that sentence should have finished that the link has failed and
the connection be closed.

Actually, ping-frequency doesn't come into play very much here.
First, it's very long, and second, the algorithm uses randomness in
its finger table searches, trying to avoid this problem.  It's more of
an issue for neighbor connections, which are tried more aggressively.
But if you're using periodic recovery, you get that automatic bit of
hysteresis.  A reactive recovery implementation probably would want to
detect and avoid rapid oscillations, but the draft doesn't go into
that detail right now.  Probably it should mention it.

Bruce
_______________________________________________
P2PSIP mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/p2psip

Reply via email to