[freenet-dev] Packet size proposal

[Michael Rogers]

Matthew Toseland wrote:
> Sure. But it will cost them. RSTs are trivial. The Golden Shield uses RSTs for 
> example, rather than remembering which streams it wants to kill. Because 
> statefully killing streams would cost many times more.

Killing, yes, but if they just want to shape the traffic then RED is 
cheap and stateless. I don't know why Comcast has decided to use RSTs 
instead of traffic shaping, but sooner or later they'll have to move to 
traffic shaping as more P2P traffic is encrypted.

> Throttling UDP 
> likewise would cause other problems: it would slow down skype dramatically, 
> alienating a lot of users, so they'd need to put more hardware in to detect 
> skype...

I'm not sure about that - reducing VoIP traffic is the second major 
selling point for these devices after reducing P2P traffic. :-)

> Classic STUNT is far more complex than UDP traversal, requires listening on 
> raw sockets (i.e. needs root), and requires using a globally reachable STUNT 
> server, which is required to send a spoofed SYNACK to each side!

STUNT has moved beyond that technique, I believe these days they're 
using simultaneous open and port prediction, both of which can be 
coordinated by a third peer so you don't need any dedicated servers or 
spoofing - it's similar to UDP hole-punching but with tighter timing.

Cheers,
Michael