For many people reinventing the wheel is a natural process of scratching the proverbial itch. It is also frequently the fastest route from the idea to the prototype to the beta. *This* is enough of a justification for creating a new protocol. PS. Having extensively dealt with H.323 and SIP in the past, I don't share your enthusiasm about SIP at all. I personally find it rather poorly designed, not to mention that it is text-based. Moreover if you consider SIP in the context of signaling protocols, the situation will be rather similar to that of BT vs. HTTP. The very same arguments you gave against BT could be applied to SIP and they would hold true. That is not to say that I like BT protocol design :) Alex _____
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Adam Fisk Sent: Friday, January 05, 2007 1:16 PM To: theory and practice of decentralized computer networks Subject: Re: [p2p-hackers] HTTP design flawed due to lack of understandingofTCP That make sense, Serguei, particularly in terms of the effect of the algorithm (not sure that was its original intent!). The nebulous purpose of it is what really gets me from the protocol design perspective. If it were some really killer feature that, say, tripled throughput, it might justify creating a new protocol, but it's just not. -Adam On 1/5/07, Serguei Osokine <[EMAIL PROTECTED]> wrote: On Friday, January 05, 2007 Adam Fisk wrote: > Tit-for-tat is basically providing incentive to keep your client > running. I would argue that it is actually a way to loosely group the similar-bandwidth hosts together in the cloud. Not entirely sure why - maybe this simplifies the selection of peers, allowing you to use what amounts to a constant constant number of them instead of having to grab more and more when their cumulative uplink bandwidth happens to be inadequate. In any case, if you are downloading, your client is already running, isn't it? Unless you're talking about "keep the honest client running", as opposed to the hypothetical cheating leecher client, which is really made pretty impractical by tit-for-tat. (Not sure how serious this problem was to beging with, but that is a long discussion.) Best wishes - S.Osokine. 5 Jan 2007. -----Original Message----- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] <mailto:[EMAIL PROTECTED]> ]On Behalf Of Adam Fisk Sent: Friday, January 05, 2007 12:47 PM To: theory and practice of decentralized computer networks Subject: Re: [p2p-hackers] HTTP design flawed due to lack of understandingofTCP Hi Peter- I've come to this view after closer work over the last several years with the IETF and more intimate experience with protocol design in implementing various IETF protocols, particularly within the SIP family. My initial forays into protocol design came from working on many different protocols on Gnutella, and some of the Gnutella protocols suffer from the same problems as BitTorrent. In a nutshell, well-architected protocols are designed to do very specific things well. This allows each protocol to evolve independently, with each protocol yielding control to others in the stack at the appropriate levels of abstraction. In SIP, this approach is readily apparent and strikingly effective, with SIP exclusively establishing sessions, leaving the Session Description Protocol (SDP) to describe the session, the MIME specifications within SDP to describe the type of media the session will handle, and with STUN and ICE handling thorny NAT traversal issues. Each protocol is independent of the others, with these discrete building blocks leading to incredible flexibility as the protocols evolve. It also allows discrete open source projects to be extremely focused in the protocols they implement. One key to these principals is to re-use protocols effectively. With everyone in the world implementing and understanding MIME, SDP can interoperate much more easily if it also uses MIME. For file transfers, HTTP is the universal standard for lots of good reasons. BitTorrent uses effectively a proprietary file transfer protocol, thereby breaking interoperability with the rest of the Internet. While BitTorrent is "open" in the sense that anyone can implement it, it's almost worse than being a closed protocol because it doesn't fit in with any of the very well-designed other protocols out there. It would never have a chance to interoperate with, say, SIP or XMPP because it just implements everything as it damn well pleases. I say the features of BitTorrent don't come anywhere near justifying this because the primary reason for breaking HTTP is tit-for-tat support. Tit-for-tat is basically providing incentive to keep your client running. That's more or less fine, but that piece should not be coupled to file transfers. At a protocol design level, that's just insanity. It also comes at a tremendous cost. Every web server on the planet is now an invalid source for a file! Excluding the most powerful computers on the Internet from the distribution system doesn't seem like a sound design decision, particularly for the poorly conceived tit-for-tat justification described above. I actually have lots of other issues with BitTorrent, but the protocol layering issue might be the biggest. -Adam On 1/5/07, Peter K Chan <[EMAIL PROTECTED]> wrote: Adam, Your assessment of BitTorrent caught my attention. How is BitTorrent "breaking interoperability with the rest of the Internet?" Why is it that the unique features of BT "don't come anywhere near justifying" it? Peter _______________________________________________ p2p-hackers mailing list [email protected] http://lists.zooko.com/mailman/listinfo/p2p-hackers
_______________________________________________ p2p-hackers mailing list [email protected] http://lists.zooko.com/mailman/listinfo/p2p-hackers
