Re: Generation of traffic in "settled" peering arrangement
At 01:49 PM 08/24/1998 -0700, Patrick Greenwell wrote:
Are you saying that someone should be forced to pay for the privilege of offering something for free to your customers? Things that your customers, who I number among are requesting?
There's no way to know what is being offered "for free", "in exchange for ads", "due to a credit card transaction", etc. It's all just traffic. Customers who receive traffic currently bear some of the costs and the sending customer bears some of the costs. In the case of an off-net sender with shortest-exit routing and no offsetting traffic in the other direction, the receiving customer ends up bearing all of the costs. /John
On Mon, 24 Aug 1998, John Curran wrote:
At 01:49 PM 08/24/1998 -0700, Patrick Greenwell wrote:
Are you saying that someone should be forced to pay for the privilege of offering something for free to your customers? Things that your customers, who I number among are requesting?
There's no way to know what is being offered "for free", "in exchange for ads", "due to a credit card transaction", etc.
Actually, since we were discussing sites such as sunsite, I think that we at least have an idea. If we go to a settlement-based model, one can only wonder what will happen to freeware/shareware sites.
It's all just traffic.
Sure it is. Traffic that the *receiver* is requesting.
Customers who receive traffic currently bear some of the costs and the sending customer bears some of the costs. In the case of an off-net sender with shortest-exit routing and no offsetting traffic in the other direction, the receiving customer ends up bearing all of the costs.
Well, my understanding is (and someone correct me if I am wrong) in at least the case of Exodus, they aren't using closest-exit. I can completely understand requiring peers not use closest-exit. That seems somewhat reasonable. But it is still the receiver that is initiating the transfer in these cases and "causing" the asymmetry. And the receiver, someone like me, is paying for a bidirectional link to BBN already. I'm already paying to receive the data, yet what you are proposing to do has the sender paying as well. I haven't seen anything in these recent discussions to suggest that BBN would be offering me a discount on inbound traffic since now the sender would be paying for it. /\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\ Patrick Greenwell (800) 299-1288 v Systems Administrator (925) 377-1212 v NameSecure (925) 377-1414 f \/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/
John Curran wrote:
Customers who receive traffic currently bear some of the costs and the sending customer bears some of the costs. In the case of an off-net sender with shortest-exit routing and no offsetting traffic in the other direction, the receiving customer ends up bearing all of the costs.
I guess 'all the cost' means most of the cost, and 'no offsetting traffic' means 'not much offsetting traffic'. However, is the real problem here the traffic assymetry, or the fact that all of the traffic is coming from one geographic location? If it is the former, then there isn't much of a solution except to merge with a network that sucks a huge amount of traffic. However, if it is the latter, then wouldn't content distribution fix it? I know many web farms offer distributed servers to their customers as a type of premium service. However, since in this case it benefits all parties involved, it seems to me that it might make sense to offer this service to huge web sites at little or no additional cost. Alec -- +-------------------------------------+----------------------------------+ |Alec H. Peterson - ahp@hilander.com | Lead Network Architect | |http://www.hilander.com | Erols Internet - an RCN Company | +-------------------------------------+----------------------------------+
At 08:19 PM 08/24/1998 -0400, Alec H. Peterson wrote:
John Curran wrote:
Customers who receive traffic currently bear some of the costs and the sending customer bears some of the costs. In the case of an off-net sender with shortest-exit routing and no offsetting traffic in the other direction, the receiving customer ends up bearing all of the costs.
I guess 'all the cost' means most of the cost, and 'no offsetting traffic' means 'not much offsetting traffic'.
(yes)
However, is the real problem here the traffic assymetry, or the fact that all of the traffic is coming from one geographic location?
Traffic Assymetry of traffic which requires (for lack of better term) "transport" to the destination. In the case of longest-exit routing, the traffic received doesn't require much transit, nor in the case of distributed content.
If it is the former, then there isn't much of a solution except to merge with a network that sucks a huge amount of traffic.
Presuming shortest-exit, correct.
However, if it is the latter, then wouldn't content distribution fix it? I know many web farms offer distributed servers to their customers as a type of premium service. However, since in this case it benefits all parties involved, it seems to me that it might make sense to offer this service to huge web sites at little or no additional cost.
It's difficult to "know" whether the distribution is actually working and resulting in interconnecti traffic which is local, but otherwise, yes. /John
John Curran wrote:
It's difficult to "know" whether the distribution is actually working and resulting in interconnecti traffic which is local, but otherwise, yes.
Come on John, use your imagination. There are creative ways one can use things like cisco flow-stats to figure out how much traffic is traversing your cross-country links. As long as you too are doing closest exit routing then it is possible to get some very good distribution of server load based on the destination's location on your network. I spoke about one possible solution at NANOG 11, http://www.hilander.com/nanog11. Of course, this requires the web farms to actually deploy their huge clients using such a content distribution scheme. Alec -- +-------------------------------------+----------------------------------+ |Alec H. Peterson - ahp@hilander.com | Lead Network Architect | |http://www.hilander.com | Erols Internet - an RCN Company | +-------------------------------------+----------------------------------+
participants (3)
-
Alec H. Peterson -
John Curran -
Patrick Greenwell