> -----Original Message----- > From: [EMAIL PROTECTED] > [mailto:[EMAIL PROTECTED] On Behalf Of P. > Oscar Boykin > Sent: Wednesday, January 23, 2008 11:20 AM > To: theory and practice of decentralized computer networks > Subject: Re: [p2p-hackers] IP-over-P2P overlay with NAT traversal > > Jiang, > > I'm on of the co-authors of Brunet and a colleague of Renato's. I > thought I'd comment.
[snip] > > 2. How do you set the timeout value for the retransmission timer? > > Pretty standard stuff here: we look at the mean and std. dev. of round > trip times. When the time we have waited exceeds 5 std dev (I think > it's set at five now), we retransmit. We allow 6 retransmissions > before signaling an error. These figures have been set in an ad-hoc > fashion based on planetlab testing. They can probably be improved. > > Also, we abstract the concept of a Sender in our system. So, lots of > objects are Senders, which means they can (attempt) to deliver some > datagram to some remote place. The one hop connections in our system > are called Edges. For now, we use different statistics for edges > (which never involve multi-hop routing) than the statistics for > routing on the overlay (which may involve many hops). > > Lastly, we set a minimum value on the timer to be 2 seconds since we > have seen that processing is very bursty on planetlab, and often you > get no CPU time for a second or more on loaded nodes. Just out of curiosity - have you considered using standard TCP over your datagram layer for the purpose of reliable delivery ? Alex _______________________________________________ p2p-hackers mailing list [email protected] http://lists.zooko.com/mailman/listinfo/p2p-hackers
