Confusing title, isn't it? Well, that's how it's been the past few days - many have commented on the "Microsoft response to BitTorrent" aka Avalanche but rare few seem to understand.

John Dvorak (of PCMag fame) thinks it's The Scheme to Discredit BitTorrent (God I hate PCMag URLs, just look at this one!). I don't see how a research paper can discredit a working implementation (many platforms, all clients successfully interoperate). This appears to be usual anti-Microsoft FUD and trying to get some attention.

Bram Cohen (the author of BitTorrent) thinks that paper is complete garbage - note, the paper but not the work of Microsoft researchers. BitTorrent is Bram's creation and it's somewhat understandable that he feels bad about some false statements in the paper (more details below).

More conspiracy theories can be found at the i-kew: Avalanche, vapour and dirty tricks.  This just makes no sense - in short, author argues that Microsoft needs good P2P technology to help ease the load off of IM servers. Repeat after me: P2P is not (yet) suitable for distributing real-time content. That means no audio/video chat, streaming movies and no IM. It's just not designed for such scenario. If you want to upload 40GB of videos to a few hundred thousand people with your pathetic 128Kb ADSL upload link, no problem.

Having implemented (on Microsoft Windows platform using Microsoft .NET) and thus understood BitTorrent protocol here's what I think... The presentation of Microsoft researchers, especially in the PowerPoint form, feels a bit dated and is simply incorrect at a few places:

1. They mention successful distribution of 1.77GB RedHat ISO image; that's small today, just for fun I got 2.6GB Fedora Core ISO image over BitTorrent yesterday in under hour and a half
2. One slide mentions how clients sometimes request pieces of the file from servers - this never happens, neither with BitTorrent, nor with similar system that (to my knowledge) came before BitTorrent called eDonkey
3. "If there are many users, deciding which is the best piece to download can be very hard" - nonsense :) experiments (and common sense) show that it's very efficient to always get the rarest piece first
4. "Solutions that require to have full knowledge of who has what are not scalable" - not quite correct; while I agree that getting rid of bookkeeping "who has what" might improve the scalability, practice shows that keeping clients up to date with this information is trivially cheap and does not hamper scalability
5. "Rare blocks are hard to obtain" - having in mind "rarest piece gets downloaded first" as a default strategy of most BitTorrent implementations, I simply do not see how this can hold

But before I dismiss the slides as a dirty trick, the scheme to discredit BitTorrent or simply delusional writings of a madman ;), let me tell you that it raises some valid points. If we could improve P2P network resiliency with a (computationally reasonably cheap) scheme similar to what Avalanche is offering it would be great for all P2P content distributors.

It's a pity these two technologies (Avalanche and BitTorrent) are seen as competing ones, not something that could as a joint effort produce even better P2P network. Don't forget that Avalanche has not been implemented yet while BitTorrent has been implemented to death on many platforms and using many languages - have this in mind while comparing (if you must compare, that is).