dslreports logo
    All Forums Hot Topics Gallery


how-to block ads

Search Topic:
share rss forum feed

Philadelphia, PA

1 recommendation

Recent Comcast P4P Trial Results (re P2P)

Of potential interest to P2P users...

In a recent post @ http://www.dslreports.com/forum/r21309318- I hinted that since the spring a number of engineers have been working on P2P-related things, and that some information would start to trickle out soon. Well, soon is now for one aspect of this at least.

We concluded a P4P trial over the summer, and discussed this at the IETF meeting in July. At that time we promised to release more details ("the trial results were really good" tends not to cut it with engineers), which we've just done in http://www.ietf.org/internet-drafts/draft-livingood-woundy-p4p-experiences-02.txt. (Also picked up on ars technica @ http://arstechnica.com/news.ars/post/20081103-comcastic-p4p-trial-shows-80-speed-boost-for-p2p-downloads.html)

We'll be presenting this in a couple of weeks at the next IETF meeting, and we'll make that presentation available at http://www.comcast.net/networkmanagement/ around that time.


It's nice to see Comcast taking an active part in growing with the Internet.

Personal attacks in 3... 2... 1...

Boynton Beach, FL
reply to jlivingood
I've only read the draft once, so I may not be picking up on things that I should be from the document, but a few initial questions if you're willing...

1) iTracker - Initially while reading the document, iTracker equaled BitTorrent tracker in my mind up until about the end of page 7 where my impression changed. It sounds like the iTracker is a separate ISP provided mechanism that may or may not (depending on the P2P style) be used in conjunction with a conventional BitTorrent tracker located possibly outside of the ISPs network?

2) If the iTracker is a separate mechanism and implementation, how flexible will it be towards other P2P activities and possibly other non-P2P related activities. The example that comes to mind is for traditional distribution where a user is prompted to pick a HTTP mirror site "closest" to them. Realistically, most users only have geographical references to go by when selecting a mirror while this would seem to imply a bit of code could give them the top three best choices for them based on their current location.

3) Again, assuming it is a separate mechanism and there is some sort of API that a client application would use to communicate with the service. Also ignoring any complexities used in deriving the result, from the application perspective would it be as simple as the application asking: What is the cost to ##.##.##.## from me? The result being something like a number from 0-100 with zero being "real close" to 100 being "a world away"? Is the cost provided from an inbound, outbound, both or not really either perspective?

4) Regarding configuration, I know from the draft the intent is to have users opt-in to using the service, but is it expected the user will need to configure their application(s) for a given ISP such as Comcast or will there be a way for a client application (conventional or web based) to auto-magically suggest using the service based on detecting the availability of the resource? Probably a bad example, but I guess what I am asking is will the user need to enter something like "itracker.comcast.net" in to their application(s) or will it be able through a mechanism such as an anycast address be able to detect availability automatically? If it is a manual configuration item, I guess my concern would be laptop users who use more than one ISP on a frequent basis (forgetting to reconfigure) and if it is completely manual many people might not give it a try or know that the service is provided by their ISP.

5) IPv6 support?

Again, I've only read it completely once, but to me it sounds as if its a service that would be provided by an ISP to allow a user's application to retrieve data representing a cost or weight to any another peer on or off that user's ISP network. That cost or weight in turn can be used by the user's application to prioritize or otherwise favor certain peers with the presumption that they will receive a benefit such as better connectivity possibly resulting in better throughput for both users.

This is just my opinion and my initial reaction, but, the name... Unless I am wrong and the iTracker is just a more clever BitTorrent tracker in which case it might make sense, I would recommend staying as far away from the word tracker as I could for a service that would be implemented by an ISP. I know that probably sounds a bit irrational or paranoid and maybe it is a bit, but that is just my opinion.

Media, PA
I would recommend looking at the following website that describes how the field tests worked:


This should help you understand how the trackers worked.

Boynton Beach, FL
Thank you for the link, it does help give a bit more information and background to help put things in to perspective.