NNSquad - Network Neutrality Squad
[ NNSquad ] Re: Keith Dawson: uTorrent Responds Regarding UDP Usage
George Ou expressed concern over the upcoming switch from TCP to UDP in Bit Torrent. Specifically alleging: . UDP used mostly by "self-throttling" low-bandwidth apps.
BDG: The new BT protocol _claims_ to be self throttling, if not low-bandwidth. If implementation matches the claims, it will throttle back _before_ other applications do.
. Routers drop TCP traffic before UDP traffic because of assumptions about who uses what.
When we think about all the ramifications of a major change like this to a protocol that makes up a significant portion of all Internet traffic, it becomes clear that maybe BitTorrent didn't think this through enough.
One of the things about capitalism is that it assumes everybody will do their best _for themselves_ (within a set of rules to deter cheating) and that the market will sort it out. The Internet is very capitalist -- it has a set of underlying rules (the RFCs) and everybody is free to innovate within that loose framework. The history so far suggests that the net (ISPs, Tier 1 carriers, end users, and application developers) has done a very good job of adapting to changes as they come along. Usually this occurs quite rapidly. The only exception I can think of is the "Internet meltdown" referred to in another message -- which I remember as happening in the 1990s rather than 1980s. That was a problem that couldn't be solved in software and required adding a lot of capacity. It took about 18 months before the Internet became usable in the middle of the day again. (Even then, things worked fine from about 3PM to 10 AM)
In any case, Ou's analysis ignore the 8 bits dedicated to Type of Service in IPv4 packets (which includes UDP). http://tools.ietf.org/html/rfc791 (search for "Type of Service"). Proper use of this will allow BT to specify a lower precedence and greater tolerance for delayed and lost packets, while VoIP can specify a high precedence and lower tolerance for delay & packet loss. Gaming applications and DNS could specify intermediate precedence and their own relative tolerance. If BT plays fair, then there should be no problem *and* ISPs will be able to use the ToS bits to determine which packets to discard, *without* Deep Packet Inspection and other RFC violations.
If the uBT press release is to be trusted, they do intend to play fair with the system, and do their best to "get out of the way" at peak congestion times. If they follow through, it would be as if the trucking associations got their drivers to agree to pull over to the side of the road during rush hour and only start their engines again when traffic eased up.
There is, of course, no guarantee that BT will do what they say. If they don't, ISPs can still manage traffic by the method most NT supporters have been advocating: count packets and bytes to and from each user. You get an allocation of so many packets & bits at priority 0, so many at prio 1, etc. Exceed your allocation and your traffic will automatically be downgraded to a priority where you havent' exceeded your allocation (priority 7 would be unlimited, but also would be the first packets dropped when the network becomes congested).
This _is_ a little more difficult than just looking at individual packets -- you need a memory and you may need communication between routers, because packets to/frome a given endpoint may go by different routes. But the ISPs have now had something like 18 months to respond in a sensible way. I suspect they won't do so until forced, and then they will scream, "we can't possibly meet that deadline". Sound like the US car manufacturers just before the Japanese nearly put them out of business in the 70s? (Or 2008-2009 when some of the Big Three may actually go out of business while Toyota and Nissan are building new plants _in the US_?)