4 ms·

This entire model is duplicitous on Comcast's part. They complain when content providers dump too much content on their network. Never mind the fact that Comc

This entire model is duplicitous on Comcast's part. They complain when content providers dump too much content on their network. Never mind the fact that Comcast's entire infrastructure is built to deliver lots of downstream traffic. If, for example, a lot of Comcast customers started seeding BitTorrent traffic to L3 customers, generating upstream traffic, Comcast's network would collapse because it is not designed to do this. Comcast should be thankful that their peer is sending them content directly, and that they don't have to pay transit costs from their upstream provider. This is clearly an attempt by Comcast to put up a pay-wall for content from 3rd parties. I've noticed more and more cable company manipulation of CDNs. For example, nothing scientific, but a few days out of each month, Cablevision, my ISP in Connecticut, starts randomly delaying traffic to Akamai CDN which serves Hulu content. It is difficult for me to scientifically prove anything is wrong, however, when the entire Internet is working flawlessly and a few random Akamai CDN content addresses, which just happen to have Hulu content, stop working, or are throttled, something stinks.

davidu 16y ago
I guess we disagree. I don't find anything you wrote to be very factual. Most networks charge their neighbors -- be them customers or peers. That means when one neighbor sends traffic to another neighbor via the ISP, they both usually pay. If you call that duplicitous, then every network is duplicitous. Also -- Comcast's network is symmetrical to the head-ends. It's true the head-ends are configured to allow more spectrum to be used for download than upload, but that's easily changed if people's surfing habits changed. I imagine in the future it'll be dynamic, per customer (if we even continue to use this model for cable modems, which I think DOCSIS is starting to move away from).
- illumin8 16y ago
  Most peering agreements don't work like ISP customer agreements. Peering agreements are usually entirely agreements NOT to charge for traffic which terminates directly on your network. Transit agreements are more like typical ISP customer relationships, where I pay you to transit traffic that has an endpoint on some other network. Ignoring the fact that the entire DOCSIS standard was designed around downstream distribution of content, and simply would not function well if you try to reverse the flow, how do you determine who pays who in your world? If I'm Comcast, do I have to pay L3 because they have content my customers want? Or should L3 pay me because I have customers they want to deliver content to? You see, it is not so cut and dried as "everybody pays for bandwidth." ISP customers pay for bandwidth because they are asking the ISP to transit traffic through their network. Peers typically don't pay because they are both endpoints.
  - davidu 16y ago
    1) How do you know what peering agreements look like? What you write is simply untrue from my experience. I peer with about 200+ networks today. Who are you? Illumin8 me please. 2) Of course DOCSIS works if you reverse the flow. You already have separated spectrum in each channel. The wider one is typically for downstream, but could easily be converted for upstream. As I said, I think in the future it'll be dynamic.
    - illumin8 16y ago
      You said: "That means when one neighbor sends traffic to another neighbor via the ISP, they both usually pay." This is true, if they are both customers of an ISP, and they are using that ISP to transit traffic to each other, they both pay. But if we have a lot of traffic destined for each other, it would make sense for us to peer and reduce our bandwidth costs from our upstream providers. DOCSIS only works in reverse if you are willing to install cable plant equipment at a customer site, which would mean a rack full of head-end gear at every house. Not feasible, and won't happen tomorrow. I used to handle BGP and peering for a tier 1 ISP and CLEC in the western US, so I've seen my fair share of peering agreements, and I've run my fair share of DS3, OC3 and OC12 circuits to peering locations. It sounds like you have a lot of credentials working for OpenDNS, so I don't mean to denigrate your experience, but OpenDNS seems more like a bandwidth customer and less like a provider/peer. I have a lot of experience with cable plant equipment because we provided cable modems and DOCSIS equipment to a lot of small business/enterprise customers in the late 90s/early 2000s.

davidu 16y ago

I guess we disagree. I don't find anything you wrote to be very factual. Most networks charge their neighbors -- be them customers or peers. That means when one neighbor sends traffic to another neighbor via the ISP, they both usually pay. If you call that duplicitous, then every network is duplicitous. Also -- Comcast's network is symmetrical to the head-ends. It's true the head-ends are configured to allow more spectrum to be used for download than upload, but that's easily changed if people's surfing habits changed. I imagine in the future it'll be dynamic, per customer (if we even continue to use this model for cable modems, which I think DOCSIS is starting to move away from).

illumin8 16y ago
Most peering agreements don't work like ISP customer agreements. Peering agreements are usually entirely agreements NOT to charge for traffic which terminates directly on your network. Transit agreements are more like typical ISP customer relationships, where I pay you to transit traffic that has an endpoint on some other network. Ignoring the fact that the entire DOCSIS standard was designed around downstream distribution of content, and simply would not function well if you try to reverse the flow, how do you determine who pays who in your world? If I'm Comcast, do I have to pay L3 because they have content my customers want? Or should L3 pay me because I have customers they want to deliver content to? You see, it is not so cut and dried as "everybody pays for bandwidth." ISP customers pay for bandwidth because they are asking the ISP to transit traffic through their network. Peers typically don't pay because they are both endpoints.
- davidu 16y ago
  1) How do you know what peering agreements look like? What you write is simply untrue from my experience. I peer with about 200+ networks today. Who are you? Illumin8 me please. 2) Of course DOCSIS works if you reverse the flow. You already have separated spectrum in each channel. The wider one is typically for downstream, but could easily be converted for upstream. As I said, I think in the future it'll be dynamic.
  - illumin8 16y ago
    You said: "That means when one neighbor sends traffic to another neighbor via the ISP, they both usually pay." This is true, if they are both customers of an ISP, and they are using that ISP to transit traffic to each other, they both pay. But if we have a lot of traffic destined for each other, it would make sense for us to peer and reduce our bandwidth costs from our upstream providers. DOCSIS only works in reverse if you are willing to install cable plant equipment at a customer site, which would mean a rack full of head-end gear at every house. Not feasible, and won't happen tomorrow. I used to handle BGP and peering for a tier 1 ISP and CLEC in the western US, so I've seen my fair share of peering agreements, and I've run my fair share of DS3, OC3 and OC12 circuits to peering locations. It sounds like you have a lot of credentials working for OpenDNS, so I don't mean to denigrate your experience, but OpenDNS seems more like a bandwidth customer and less like a provider/peer. I have a lot of experience with cable plant equipment because we provided cable modems and DOCSIS equipment to a lot of small business/enterprise customers in the late 90s/early 2000s.