* Matthew Toseland <toad at amphibian.dyndns.org> [2006-04-26 20:40:16]:
> On Thu, Apr 27, 2006 at 04:43:19AM +0200, Florent Daigni?re (NextGen$) wrote: > > * Matthew Toseland <toad at amphibian.dyndns.org> [2006-04-26 19:01:30]: > > > > > On Wed, Apr 26, 2006 at 02:31:00PM +0000, NextGen$ wrote: > > > > * Matthew Toseland <toad at amphibian.dyndns.org> [2006-04-26 13:26:33]: > > > > > > > > > On Wed, Apr 26, 2006 at 11:19:23AM +0000, NextGen$ wrote: > > > > > > > > > > > > imho Up&p and stun are useless without multi-homming support. > > > > > > > > > > Huh? What exactly do you think is necessary prior to STUN/UP&P? > > > > > > > > Having references with multiple contact points (an "official" signed > > > > reference with possibly several addresses and an unofficial one with ip > > > > gathered from up&p, other peers, last known, ...)... And maybe the > > > > possibility within the node to have different listeners bound to > > > > different > > > > sockets. > > > > > > We already have the possibility of multiple IPs in the ref, and a > > > separate last-detected IP. The IP detected from other peers is just used > > > in the overall detection algorithm though; we will only publish one IP > > > address at present, and we will only use one at a time to talk to a > > > given node. I'm not sure exactly what you are suggesting we need. > > > > Imho we need to publish every known and valid ip addresses, even local > > ones ; otherwise nodes on the same lan won't be able to connect (to both > > outside and internal peers). > > Isn't that a security risk? Well, not to darknet peers I suppose? > I don't see it as a security risk. Sending one handshake packet once a while isn't a security problem imho. > Anyway we don't want to try such addresses unless we have a good reason > to believe they will work e.g. if we have the same external IP detected > through STUN ? ... whereas sending packets to an external well known 3rd party peer is ;) Even if we bundle an ip-list of stun servers... a dns name... It will be a convenient way to harvest. > > True multi-homing as you suggest would be fairly easy though, it's not a > big deal. I'm not saying it's hard to do :) Just that it's pointless to do the rest without the basis ;) NextGen$
