Live data from Hacker News

Changing the game with μTP

blog.bittorrent.com

1–10 of 20 posts

Re: Changing the game with μTP

#2
This new protocol uses packet latency based throttling instead of the packet loss based throttling of TCP.

I don't think the article documents any details. Maybe I missed a link.

Re: Changing the game with μTP

#3
post #2

This new protocol uses packet latency based throttling instead of the packet loss based throttling of TCP. I don't think the article documents any details. Maybe I missed a link.

The article doesn't mention details but in the comments, Simon links to the IETF working group on this: http://www.ietf.org/dyn/wg/charter/ledbat-charter.html

Re: Changing the game with μTP

#4
What's unsaid here is that uTP is an undocumented protocol that isn't available to other BitTorrent implementations like Azureus/Vuze, potentially making them second-class citizens.

Re: Changing the game with μTP

#5
post #2

This new protocol uses packet latency based throttling instead of the packet loss based throttling of TCP. I don't think the article documents any details. Maybe I missed a link.

Latency based congestion control already exists for TCP[1].

So my question is, why wouldn't you get the same effect by running BT on top of a network stack using Vegas?

Someone flash the tptacek signal!

[1] http://en.wikipedia.org/wiki/TCP_Vegas

Re: Changing the game with μTP

#6
post #5
post #2

This new protocol uses packet latency based throttling instead of the packet loss based throttling of TCP. I don't think the article documents any details. Maybe I missed a link.

Latency based congestion control already exists for TCP[1]. So my question is, why wouldn't you get the same effect by running BT on top of a network stack using Vegas? Someone flash the tptacek signal! [1] http://en.wikipedia.org/wiki/TCP_Vegas

why wouldn't you get the same effect by running BT on top of a network stack using Vegas?

You probably would, but Windows doesn't support Vegas.

http://www.ietf.org/mail-archive/web/p2pi/current/msg00052.h...

Re: Changing the game with μTP

#7
post #4

What's unsaid here is that uTP is an undocumented protocol that isn't available to other BitTorrent implementations like Azureus/Vuze, potentially making them second-class citizens.

from the bittorrent.com comments, a quote from a high-up at BT:

Simon Morris

Posted October 8, 2009 at 5:52 pm

As a matter of fact µTP is indeed a replacement for TCP, so we think the name is appropriate. You can check out standardization work which we’re chairing at the IETF called the LEDBAT working group

Re: Changing the game with μTP

#8
post #5
post #2

This new protocol uses packet latency based throttling instead of the packet loss based throttling of TCP. I don't think the article documents any details. Maybe I missed a link.

Latency based congestion control already exists for TCP[1]. So my question is, why wouldn't you get the same effect by running BT on top of a network stack using Vegas? Someone flash the tptacek signal! [1] http://en.wikipedia.org/wiki/TCP_Vegas

It appears to me that if you use Vegas on your Linux machine then all of your TCP connections use Vegas, which will make you a second class citizen to all the Reno folk out there on the internet.

If you assume that your own uplink is the only bottleneck to be dealt with (say you are serving torrents over a DSL line) then this is reasonable, but I don't think I'd want to yield to all TCP users everywhere with all of my outgoing traffic.

There are also some scary comments in the notes from before it was rolled into the kernel, like the code doesn't handle route changes.

I can see why a vendor would choose something over which they have control.

Re: Changing the game with μTP

#9
post #7
post #4

What's unsaid here is that uTP is an undocumented protocol that isn't available to other BitTorrent implementations like Azureus/Vuze, potentially making them second-class citizens.

from the bittorrent.com comments, a quote from a high-up at BT: Simon Morris Posted October 8, 2009 at 5:52 pm As a matter of fact µTP is indeed a replacement for TCP, so we think the name is appropriate. You can check out standardization work which we’re chairing at the IETF called the LEDBAT working group

LEDBAT is only standardizing a small subset of uTP:

"Framing is not on the WG's charter."

http://www.ietf.org/mail-archive/web/ledbat/current/msg00117...

Re: Changing the game with μTP

#10
post #4

What's unsaid here is that uTP is an undocumented protocol that isn't available to other BitTorrent implementations like Azureus/Vuze, potentially making them second-class citizens.

That should be a nonissue as long as the protocol is backwards compatible.
Post reply on HN