Today was the deadline for Comcast to tell the FCC how its existing congestion management system works, as well as how its “protocol agnostic” replacement is going to work. To the dismay of some critics, they’ve done just that in a filing that was hand-delivered as well as electronically filed today. It will be posted to the Comcast web site shortly.
The filing corrects some of the false allegations made by critics with respect to privacy, making it very clear that the existing system simply inspects protocol headers (“envelopes”) and not personal data. David Reed in particular got himself worked into a tizzy over the idea that Comcast was deciding which streams to delay based on content, but this is clearly not the case. Inside the IP envelope sits a TCP envelope, and inside that sits a BitTorrent envelope. User data is inside the BitTorrent (or equivalent) envelope, and Comcast doesn’t look at it.
The current system sets a bandwidth quota for P2P, and prevents P2P as a group from crossing the threshold from this quota (about 50% of total upstream bandwidth) with new uni-directional upload (AKA, file-server-like) streams by tearing down requested new streams with the TCP Reset bit. The system is a bit heavy-handed, but reserving 50% of the network for one class of application seems pretty reasonable, given that no more than 20% of customers use P2P at all.
Nonetheless, the new system will not look at any headers, and will simply be triggered by the volume of traffic each user puts on the network and the overall congestion state of the network segment. If the segment goes over 70% utilization in the upload direction for a fifteen-minute sample period, congestion management will take effect.
In the management state, traffic volume measurement will determine which users are causing the near-congestion, and only those using high amounts of bandwidth will be managed. The way they’re going to be managed is going to raise some eyebrows, but it’s perfectly consistent with the FCC’s order.
High-traffic users – those who’ve used over 70% of their account’s limit for the last fifteen minutes – will have all of their traffic de-prioritized for the next fifteen minutes. While de-prioritized, they still have access to the network, but only after the conforming users have transmitted their packets. So instead of bidding on the first 70% of network bandwidth, they’ll essentially bid on the 30% that remains. This will be a bummer for people who are banging out files as fast as they can only to have a Skype call come in. Even if they stop BitTorrent, the first fifteen minutes of Skyping are going to be rough. A more pleasant approach would be to let excessive users out of QoS jail with credit for good behavior – if their utilization drops to Skype level, let them out in a few seconds, because it’s clear they’ve turned off their file sharing program. This may be easier said than done, and it may raise the ire of Kevin Martin, given how irrational he is with this anti-cable vendetta.
The user can prevent this situation from arising, of course, if he wants to. All he has to do is set the upload and download limits in BitTorrent low enough that he doesn’t consume enough bandwidth to land in the “heavy user” classification and he won’t have to put up with bad VoIP quality. I predict that P2P applications and home gateways are going to incorporate controls to enforce “Comcast friendly” operation to prevent de-prioritization. There are other more refined approaches to this problem, of course.
At the end of the day, Comcast’s fifteen/fifteen system provides users with the incentive to control their own bandwidth appetites, which makes it an “end-to-end” solution. The neutralitarians should be happy about that, but it remains to be seen how they’re going to react.
It looks pretty cool to me.
UPDATE: Comcast-hater Nate Anderson tries to explain the system at Ars Technica. He has some of it right, but doesn’t seem to appreciate any of its implications. While the new system will not look at protocol headers (the evil “Deep Packet Inspection” that gets network neophytes and cranks so excited) , and it won’t use TCP Resets, that doesn’t mean that P2P won’t be throttled; it will.
That’s simply because P2P contributes most of the load on residential networks. So if you throttle the heaviest users, you’re in effect throttling the heaviest P2P users, because the set of heavy users and the set of heavy P2P users is the same set. So the “disparate impact” will remain even though the “disparate treatment” will end.
But the FCC has to like it, because it conforms to all of Kevin Martin’s rabbit-out-the-hat rules. The equipment Comcast had had to purchase for this exercise in aesthetic reform will have utility down the road, but for now it’s simply a tax imposed by out-of-control regulators.
