Is Comcast (still, really) blocking BitTorrent?

The neutralists are touting a study by a German research center alleging Comcast blocking of BitTorrent at all hours of the day:

The Max Planck Institute has released a new survey of worldwide BitTorrent traffic finding that Comcast and Cox are the chief offenders for throttling traffic, and that they block at all hours of the day and night.

The study is here.

The following statement is attributed to Gigi B. Sohn, president and co-founder of Public Knowledge: “…These results lead us to three conclusions. First, the largest cable companies were doing more blocking than they have admitted to Congress or to the FCC. Second, other cable companies, and all telephone companies, can manage their networks without the need for blocking the traffic of customers. Finally, the fact that the blocking goes on all the time should tell the Commission that it needs to act soon to prevent the practice.”

Like a good engineer, I tried to measure my Comcast link using the German tool, and got this (click on the image to enlarge:)

All Systems Busy

This tells me that the test system has some problems, as the server has been busy for several hours. Feel free to try it yourself.

I know my BitTorrent isn’t blocked on my Comcast connection because I tested it in real-world conditions: yesterday, I successfully downloaded Fedora Linux version 9, over Comcast using BitTorrent, at full speed and with no interruptions. I then seeded it for several hours before shutting down Azureus (the Linux client for Vuze’s implementation of BitTorrent) and just restarted it an hour ago. So I’m now seeding without incident (click to enlarge:)

Seeding like craze over Comcast

Pardon me if I’m skeptical of Ms. Sohn’s charges. It may be that the Germans have found something, but I can’t confirm it; perhaps we need a better instrument.

The Germans are apparently measuring TCP-level behavior, the thing that CMU professor Jon Peha and the infamous David Reed have measured and commented upon. The problem with this analysis is that Vuze’s BitTorrent uses 50 TCP streams by default, continually testing them for the 4 best and making these 4 active at any given time for any given piece of content. So while some of the 50 are torn down by Sandvine management systems, they’re quickly replaced and you retain 4 active upstream TCP connections most of the time.

The test that should be done would simply record the seeding rate of popular BitTorrent sessions like mine. When you do that – test the overall contribution to the swarm, not the 50 TCP connections individually – you find that BitTorrent works quite well on cable networks, contrary to the claims that the neutrality regulation advocates are making.

The German tool simply doesn’t produce meaningful data, and the advocacy groups touting it (Free Press is another) it are deliberately misleading the public.

UPDATE: Peter Svensson, Free Press’ personal reporter, has written the predictable story about the evils of the BitTorrent blocking uncovered by the alleged research organization in Germany. It’s predictable because Svensson ran the gamut of interviews with ISP critics, but didn’t talk to anyone who might shed light on the study. In fact, reducing the number of active TCP connections from a given BitTorrent user often increases his throughput, just like metering lights on freeway on-ramps increase freeway traffic speeds. This is only one of the critical points that these amateur network engineers don’t get.

13 thoughts on “Is Comcast (still, really) blocking BitTorrent?”

  1. You obviously haven’t been reading about what Comcast has been saying. They first DENIED blocking, then changed their story to, “we only delay”. If they were not doing anything why change their story??? They changed their story because they were INDEED blocking and had to scramble and are still scrambling to put out a fire they created themselves. I mean seriously, do you REALLY believe that these huge corporations have OUR best interest at heart?? Knowledge is power, and those who control knowledge have lots of power.

  2. As I said, I simply tested my Comcast link and reported the results. My tests show that Comcast does not prevent me from using BitTorrent. What Comcast says or doesn’t say is irrelevant, what matters to me is what actually happens on the network.

    My testing shows that Comcast does not “block” BitTorrent from or to my house.

  3. Sorry Rich you need to read more about an issue before blaggin’ about it.

    They are blocking only at certain CO’s, so you may not be on a Sandvine infected connection. The test’s results show that half the traffic is being blocked, but not blocked totally. Which makes sense if you’ve read into Sandvine and the service they sell. The idea of Sandvine is to corrupt the quality of the traffic. You’ll get your torrent, while still being blocked, and degrading everyone else’s experience.

    If the peers are numerous, on bittorrent friendly ISPs, and not total leeches, it wont be a problem. The torrent will live a long and happy life, and no one goes off to complain to the FCC about interference to their paid service. But what if two people in an area with Sandvine hardware installed, do a peer to peer transfer both ways? The result is both clients spinning next to nothing.

    The hardware is still installed, and active, all the friggin time. Contrary to what they’ve told the public, and the FCC. Multiple times. Thats the point. Their original story was a lie, its revised story was another lie, what they told the FCC is a lie, and their current spin on these results is just a restated lie. Bullocks to your test, the only result of it is that we know that particular Fedora torrent is healthy. I prefer Debian anyways.

  4. Jim, Comcast didn’t change their message; they always maintain that they’re delaying BitTorrent.

    This report only confirms that individual TCP connections are being blocked; it’s being misreported as BitTorrent being blocked. BitTorrent uses dozens of TCP streams so blocking a few of those streams has the effect of slowing down BitTorrent thus “delay”. Furthermore, as TCP streams are broken, BitTorrent will naturally open up more connections automatically within seconds of connections being broken.

  5. Comcast has always said that they only prune TCP connections for BitTorrent in seeding mode, and only then when the segment is highly loaded. The effect of this is to take bandwidth from BT seeders and re-distribute it to other applications. The main beneficiary of this is actually BitTorrent itself, in the typical sharing mode. So it’s simply a matter of throttling BT in one condition to aid BT in another.

    The “study” by the Germans didn’t measure BT throughput, and that’s the important statistic.

  6. Comcast would be 100% justified if it DID block BitTorrent, because BitTorrent and other P2P programs violate its terms of service and take its bandwidth without compensation. They also degrade the performance of the network for legitimate users.

  7. Two interesting points I neglected to mention in my posting just above. Firstly, if you look at the “results” page of that German “test” site, at, you will notice an interesting pattern. If you believe their data (and, of course, it really should be independently verified), the US is, once again, on the technological forefront — ahead of nearly every country in the world. It shows that nearly 25% of the tested Internet connections in the US had some form of P2P throttling in place. This is a good thing; it means that we as a nation are doing a better job of stopping network abuse, and preventing bandwidth hogs from degrading our networks, than any other country in the world except Singapore. Good for us!

    The second point I didn’t mention above is that I’ve tried to contact Peter Svensson, whom Richard dubs “Free Press’ personal reporter,” and give him accurate information about the practice of P2P throttling by ISPs. I got through once and spoke to him for a few minutes, offering him more data if he was interested. But he never contacted me, and has continued to write stories which are biased and inaccurate (and which repeat false statements made by Free Press). Therefore, Richard’s characterization appears (alas) to be quite accurate.

  8. As a good engineer, you will realise you tested exactly nothing.

    You downloaded a torrent and it got to you. You did not determine if your download had been throttled because you cannot quantify it by downloading – you need to inspect raw packets to find the resets that comcast use as their mechanism to limit it.

    You draw a false conclusion that a website that is currently ‘busy’ has problems – did it ever occur to you that the website operators, by limiting the number of concurrent tests are ensuring that any results they do obtain are not affected by overutilisation?

    Your article is what we in the professional tech world call ‘Intellectual Masturbation’ – zero in-depth analysis and conclusions drawn on false assumptions and proven through shoddy testing methodolgy.

    Comcast DID say they were doing nothing – the Network Neutrality Group that many /.’ers are involved with have verified and documented it. When caught they changed their story to ‘we’re delaying’ -> a euphamism for ‘we’re actively resetting bittorrent connections so that it takes the end use longer to download something’.

    The issue is not and never has been ‘comcast are blocking ALL bittorrent traffic’ – but they are actively interfering with it. We don’t need any other proof beyond the forged packets themselves to know that comcast are interfering.

    To use one of Mr Glass’ style of overexaggerated analogy – The community has reported a rapist and have caught the rapist on camera – while the rapist’s supporters are screaming ‘He didn’t murder anyone!’.

  9. I measured the upload and download rates, Paul. That’s data. My point is that the upload and download rates are important, but the fact that random RSTs may or may not be injected to prune excess connections is not important.

    It’s the result that matters, not the method.

  10. I just tried the Max Planck (“walk the Planck?”) test. It failed. It said:

    We are sorry, the test failed due to a problem.

    The connection to the measurement server timed out. This can happen if it takes too long to load the applet.

  11. I would like to know how they could possibly have gathered any usable data if their tester does not work.

Leave a Reply

Your email address will not be published. Required fields are marked *