NNSquad - Network Neutrality Squad
NNSquad Home Page
[ NNSquad ] Re: NY Times: Verizon offers system to improve P2P transfers
One other comment: This is interesting for a couple of factors: The files distributed are relatively few, so they should actually cache relatively well. And with FIOS, Verizon may have the same cost for upstream/downstream as everybody else, but a vastly cheaper local loop per bit. Thus P2P makes economic sense for the ISP (that is, it doesn't cost-shift TO the ISP, but actually REDUCES cost) when: Files are VERY popular: You expect 2+ users of any given file within any "local" segment area (last mile zone to a gigabit switch). Fewer files increases the odds that for the same user population, its likely to be popular. Your local loop is NOT the bottleneck, but your uplink/downlink may be. This is especailly the case for FIOS, because of the vastly different relative costs compared to say the cable modems. The P2P system is very topologically aware, so it can take advantage of the local loop spare capacity. Files are large, so the ISP doesn't want to pay the file space/cost in its HTTP cache. So this is the sort of P2P that Verizon would really like. Although they do have to be careful, because if the local loop DOES become congested, these P2P flows will significantly outcompete "normal" TCP flows, because there may be more per transfer (multiple peers) and are very low latency, both of which make a flow far more competitive in TCP's congestion control mechanisms. So there may be a need in the future for such a P2P system to have a more gentle congestion control (backoff more on drops) or active shaping to avoid these problems. But expect Comcast to initially take the other side: They have a local loop that doesn't have nearly the spare bandwidth as FIOS, AND the offerings compete directly with their on-demand services. It may come down to "How nice do we need to play with NBC" in their decision making.