[ NNSquad ] Re: Sandvine MSO Study shows device throttles P2P upload by 98%

[Phil Karn <karn@ka9q.net>]

Lauren Weinstein wrote:

>      Once Sandvine Peer-To-Peer Policy Management was enabled,
>      bandwidth utilization by P2P traffic fell to less than eight
>      per cent (8%) [from 94%] without impacting subscriber peer to
>      peer sessions.
> 
> One might ask, how is this possible?  Well, the report suggests that
> the vast bulk of P2P protocol activity is useless "chatter" and the
> like.  An interesting assertion.

If -- *IF* -- there are many nodes in a Bit Torrent (say) session, and most of
these nodes happen to peer with nodes outside the ISP and relatively few with
each other, then a box that discriminates against inter-ISP transfers could
actually improve the transfer rates of individual nodes. They'd do this by
encouraging intra-ISP transfers and making more efficient use of limited
inter-ISP capacity by not sending multiple copies of the same data over the same
links.

This does not legitimize the technique in any way. The same results -- or better
-- can be achieved much more directly and effectively and without any violence
to the end-to-end model. The ISPs just have to provide connectivity cost
information to the P2P nodes so they can do a better job of peer selection. P2P
vendors and users would have an incentive to use this information not just to
get the ISPs off their backs but also to make their transfers go faster.

It might be possible to add algorithms to Bit Torrent to deduce connectivity and
cost information (e.g., by round trip time measurements) but you can do better
with an explicit source of this information.