Bram Cohen, inventor of the BitTorrent file-sharing protocol, has launched a stinging attack on Microsoft's "Avalanche" peer-to-peer system unveiled (sort of) last week by its Cambridge (UK) researchers.
In a posting on his blog, Cohen - whose protocol is (over?)estimated to be responsible for one-third of the world's internet traffic - criticises Avalanche for about as many things as you can shake a good-sized stick at.
"Avalanche is vaporware," he begins. "It isn't a product which you can use or test with, it's a bunch of proposed algorithms. There isn't even a fleshed out network protocol. The 'experiments' they've done are simulations. It's a bad idea to give much weight to simulations, especially of something so hairy as real-world internet behavior."
Certainly, anyone who was at the presentation last week of the Avalanche concept would have come away thinking it was a here-now, working-in-the-lab product. (We certainly did.) The truth, it turns out, is that Avalanche is study of strategies and principles of how a file-swarming system could be made to work well.
In his demolition of the "vaporware" (a neat trick in itself), Cohen begins by pointing out that the online paper with the Microsoft group's criticism of his own protocol is, at best, seriously out of date because it points up problems with BitTorrent that were removed by a rewrite "back in 2001". "In their simulations [of how BitTorrent works, or doesn't] peers have 4-6 neighbors with a strong preference for 4. BitTorrent in the real world typically uses 30-50." That effectively hobbles his baby, and the simulation doesn't look at other variables - "varying transfer rate abilities or endgame mode" - so that "intentionally or not, the simulation is completely rigged against BitTorrent."
And there's more: "One thing badly missing from this paper is back-of-the-envelope calculations about all of the work necessary to implement error correction." Particularly, he says, if the file being transferred is bigger than the memory available, you'll end up thrashing your disk as you try to recombine your whole file to send it to the next peer.
He concludes that "I think that paper is complete garbage. Unfortunately it's actually one of the better academic papers on BitTorrent, because it makes some attempt, however feeble, to do an apples to apples comparison. I'd comment on academic papers more, but generally they're so bad that evaluating them does little more than go over epistemological problems with their methodology, and is honestly a waste of time."
We asked Peter Key, joint manager of the Networks and Performance group at Microsoft Cambridge, to respond. Emphasising firstly that he doesn't want to get into a slanging match with Cohen or about BitTorrent, he acknowledged that yes, Avalanche is presently a research project that isn't publicly available to test, and that it relies on simulations to compare how it would work with existing file-sharing systems.
"Of course simulations are always going to be abstract; we're trying to understand the principles here," Key said. "And we aren't trying to develop a generalised file-sharing system. Precisely what it will be used for would be up to the product group." Quite when - or if - Avalanche would appear is uncertain too.
But Key points out that one of the important problems with file-swarming (a la BitTorrent) that Avalanche aims to solve is that of "poisoned" swarms, whereby a malicious peer can upload a little bit of a file, or change part of a file, which will then be propagated to peers. That needs error correction, which Cohen admits is tough. Microsoft's Key, by contrast, says that a focus of the Avalanche project is "securing the integrity of the files". One thing you can be sure of: there's no risk of downloading a virus from vaporware. Then again, there's not much risk of downloading anything from vaporware. ®