On Sun, 1 Mar 2009, 8:08am -0800, Nicholas Weaver wrote: > Looking at the current draft, a few thoughts.... > > The usage cases may be too broad. EG, local mirror/caching discovery > for conventional CDN operation doesn't need to rely on alto, and would > probably not benefit, as the conventional CDNs have solved the problem > "good enough", and it also interacts with their OWN secret-sauce. > > My thought would be focus on the two useage cases which count: > > Bulk data P2P peer localization, and P2P hashtable localization, where > the first is about saving cost while providing the same service, while > the latter is reducing latency. Can you elaborate some more on reducing latency only for P2P hashtable? > On security considerations, any open-world P2P system per file can be > tracked from one or more nodes by creating sybils, so ALTO doesn't make > the problem easier, and any P2P system which deliberately hides this > (eg, by onion-routing data) is not going to WANT localization and > increased localization increases the power of traffic analysis. > > Thus I believe that ALTO should NOT attempt to be privacy preserving, > because the protocols that use it won't be privacy preserving anyway.
You mean no privacy preserving only for P2P clients. I have to disagree. Here are some reasons: (1) Giving ALTO Servers such information will make it possible for some parties to demand that such information be collected; this increases potential burdens on ISPs and thus may hinder ALTO deployment. (2) Sybils have limitations, e.g., limited view, large number of torrents that are hard to track, closed trackers, closed-source systems (the world is larger than BitTorrent), symbils can be banned... ALTO Servers without privacy preservation for P2P clients can help to "solve" these problems. Richard _______________________________________________ alto mailing list [email protected] https://www.ietf.org/mailman/listinfo/alto
