DISQUS

Luis · 2 years ago

Of course, this wasn't throttling, it was active breaking of the protocol. But don't let those little details spoil a good story. :)

Tom · 2 years ago

Luis is right. And even flat throttling isn't necessary: many home routers can provide dynamic QoS, prioritizing traffic that needs low-latency connections -- VoIP is a red herring. But then, that keeps the pipe fully utilized, which doesn't help Comcast's bottom line. They have to pay for that data on a metered basis once it leaves their network, after all.

Ryan Radia · 2 years ago

I don't believe Comcast has negotiated settlement-free interconnection with any Tier 1 carrier. So presumably they have to pay for bandwidth on a metered basis, as Tom suggested. And since Bittorrent is a major source of bandiwdth usage by Comcast customers, anything that can be done to reduce file sharing traffic means lower costs for Comcast.

On the other hand, testing of the methods used by Comcast to shape traffic has determined that RST packets are being forget to restrict seeding exclusively, and downloading via Bittorrent is unaffected. Comcast's focus on upstream traffic suggests their main concern is high upstream traffic causing saturation of local cable nodes.

I'd be interested to learn more about how the peering agreements of Tier 1 ISPs like Qwest, Verizon, and AT&T; deremine their management of peer to peer traffic. Because they do not pay for bandwidth on a byte-per-byte basis, Tier 1 carriers have an incentive to fill up their pipes at all times while using QoS to prioritize traffic that is latency-sensitive like VoIP or highly valued like Email and web browsing.

Barnaby · 2 years ago

I'm guessing that Richard Bennett will give George Ou a 'gold star' and a 'pat on the back' for this story...

Wes Felter · 2 years ago

Cisco claims that their DOCSIS 1.1 CMTS supports Weighted Fair Queuing on the upstream link. I don't know if other vendors support WFQ or what CMTS Comcast is using, but I am skeptical of claims that upstream QoS is impossible.

Of course, we don't even know for sure that upstream last mile contention is the problem that Comcast is attempting to solve.

Richard Bennett · 2 years ago

I'd take Cisco's claims with a grain of salt. Their version of WFQ is apparently implemented in the CMTS rather than in the CM itself, so it's post-collision.

On your main point, Wes, you're absolutelyh right that we don't know that for sure what problem Comcast is trying to solve, we're just making an educated guess that happens to fit the facts fairly well. In the final analysis, it's up to Comcast to say why they're doing what they're doing, and they haven't shown any willingness to do so. And frankly, I'm getting tired of defending them.

George Ou · 2 years ago

Wes,

There is no "queue" in the last mile since everything that collides is dropped; not queued. How do you do enhanced queuing on a network that has no queue?

Richard Bennett · 2 years ago

There certainly are queues inside the CM and the CMTS, George. The only collisions are between the bandwidth request frames generated as a result of data being queued, and not among the actual data frames themselves. But I agree with your larger point, that WFQ in a many-to-one system is difficult.