On BitTorrent and ISP Characterization

It is generally easy to argue the importance of understanding Internet broadband availability/adoption and, the more challenging tasks of, characterizing the quality of service offered by broadband providers. Independently of size and end goal, most everyone could benefit from such information, including subscribers shopping for alternative services, companies providing reliable Internet services, governments surveying availability of Internet services to their citizens and people like me, just trying to make sense of it all.

The last few years have seen a number of notable systems and services making progress toward this challenging goal including Speedtest, Netalyzr, YouTube/my_speed, SamKnows, Glasnost, and several companies willing to provide some summary of their perspective such as Akamai and Netflix. Most recently, BitTorrent Inc. has hinted (whatever that may mean) that they were considering releasing information on worldwide ISP performance (ISPreview post) 

 

Of course, I think this is an awesome idea and that BitTorrent - the system - is a nearly perfect place to do this. In my AquaLab research group we are trying just that!

I have previously argued that to be effective, ISP characterization must be done at scale, continuously and by end users (if you were at the CAIDA AIMS  workshop early this month you may have heard this). I believe it must be done at scale, to capture the diversity of available providers and their services (this is partially why I don't buy the Samknows/Ofcom and Samknows/FCC efforts). It should be done continuously, to capture dynamic changes due to management policies (e.g. traffic shaping or oversubscribed networks) and unscheduled events (i.e.  service interruptions).  And, for sure, it should be done by end users to guarantee accuracy (who else would you trust? ).

That said, unless these guys down in California have been at this for the last 10 years and cracked, in house or through secret collaborations, some very tough research problems, I don't think we should get our hopes too high. We have been trying to do this for a while and, trust me, it's a very hairy problem. Just a few issues that could get in the way include the user-specific configurations of BitTorrent clients, swarm capacity, cross traffic, home network (mis)configuration,  DNS configuration, ... and these are really just to get you started. I am sure many readers can add to this without much effort!