> -----Original Message----- > From: [EMAIL PROTECTED] > [mailto:[EMAIL PROTECTED] On Behalf Of Ted Lemon > Sent: Wednesday, August 13, 2008 12:54 PM > To: Iljitsch van Beijnum > Cc: [email protected]; [EMAIL PROTECTED] > Subject: Re: [Int-area] [BEHAVE] Inbound and outbound connections > > On Aug 13, 2008, at 12:45 PM, Iljitsch van Beijnum wrote: > > I would like to see some evidence for this. Google Maps > only generates > > some 20 sessions for me. > > Another problem with the statistic is that it assumes that all users > are surfing google web simultaneously. In practice, use > tends to be > staggered. As long as your NAT is noticing FIN and RST > packets, and > recycling ports quickly,
Caution needs to be exercised in recylcing TCP ports quickly; see draft-ananth-tsvwg-timewait-00.txt and draft-ietf-tsvwg-port-randomization-01.txt. > you can probably do just fine with > many more > subscribers per IP address than Miyakawa-san's worst-case scenario > would suggest. > > That's all out the window if they're all doing P2P, of > course, because > those connections stay up longer. But if we have IPv6 in the > network, we should really encourage people running P2P to use it, > because it should work better for them than IPv4 anyway - no NAT > traversal problems. Without a way to accept incoming TCP sessions (e.g., UPnP or some other way to ask for a publicly-routable transport address that accepts incoming TCP SYNs), BitTorrent will consider p2p clients as leeches and download speeds will suffer: keeping TCP connections up even longer. We can't just 'encourage' p2p users to move to v6 until there is content on v6 -- this is the very same problem for v6-only clients wanting to access v6 content (there isn't any -- www.google.com, www.cnn.com, www.amazon.com, are not on v6. "ipv6.google.com" != "www.google.com"). -d _______________________________________________ Int-area mailing list [email protected] https://www.ietf.org/mailman/listinfo/int-area
