Create an Account
username: password:
 
  MemeStreams Logo

MemeStreams Discussion

search


This page contains all of the posts and discussion on MemeStreams referencing the following web page: Why Richard Bennett is Wrong. You can find discussions on MemeStreams as you surf the web, even if you aren't a MemeStreams member, using the Threads Bookmarklet.

Why Richard Bennett is Wrong
by Dagmar at 3:32 pm EST, Dec 1, 2008

I just can't think of anything nice to say about the level of misinformation in this article, so I'm having to settle for saying things that aren't entirely hostile instead. To somewhat oversimplify, Richard Bennett has published an article to The Register making the claim that a recent behavioral change for uTorrent (a popular BitTorrent client) will result in the entire Internet's bandwidth being reduced by three-quarters.

We'll start with the out-and-out lie that heads up the article:

The leading BitTorrent software authors have declared war on you - and any users wanting to wring high performance out of their networks.

He then explains that this is because the authors of uTorrent have decided to make a change to the client so that it will default to UDP instead of TCP for data transfers. He goes on to compound this erroneous deduction with some, well, insanity.

By most estimates, P2P accounts for close to half of internet traffic today. When this traffic is immune to congestion control, the remaining half will stumble along at roughly a quarter of the bandwidth it has available today: half the raw bandwidth, used with half efficiency, by 95% of internet users. Oops.

There are some fundamental flaws in the argument he's trying to make here, one of the most glaring ones being that UDP is "immune to congestion control" (which he makes by proxy by quoting someone else, for those crying "but he's not the one that said it!"). This is simply untrue. UDP is no more immune to congestion control than TCP or ICMP. That is, it's not immune to actual congestion control... If your idea of "congestion control" is to spoof a disconnection request for a substantial number of the active connections going over a network segment (which in the 90's was called a "denial of service attack") then UDP would indeed be immune to this because UDP doesn't have a built-in "session" concept that can be so easily broken by an attacker, but that's not congestion control--at best that's connection harassment. (We won't go into detail right now about how badly Sandvine malfunctioned in addition to being the wrong solution.)

UDP was intended for real-time data transfers such as VoIP that typically move small amounts of data with a low tolerance for delay. [...] Bulk data transfers are supposed to use TCP, in large part because it shoulders the burden of congestion control for the internet’s end-to-end layer.

This is a more subtle derangement of the truth, but is no less untrue than the other premise Bennett makes. UDP was designed without handshaking protocols or methods of guaranteeing packet delivery so that it could facilitate very short transactions where handshaking or packet reassembly could introduce unuseful delays. There's no reason it can't be used for bulk transfers--it... [ Read More (0.4k in body) ]


 
RE: Why Richard Bennett is Wrong
by Decius at 3:51 pm EST, Dec 1, 2008

Dagmar wrote:
Bennett goes on to talk about the "Jacobson Algorithm" which is basically a mechanism by which a host that's sending data can be told to slow down the rate at which it sends. Sadly, Bennett fails to realize that just because UDP doesn't have this doesn't mean a similar mechanism can't be applied to it.

Bennett's article was obviously sensationalistic but one of the things I've heard Noteworthy say is that designing a congestion control algorithm is like designing your own cryptographic system. Unless you are a highly specialized expert you don't know enough about it to try. So, this begs the question, what is bittorrent's congestion control algorithm and is it any good?


 
 
Powered By Industrial Memetics