[In the message entitled "Re: BBN/GTEI" on Aug 21, 13:48, Michael Dillon writes:]
If such scalable peering already existed, I'm convinced that the current situation between Exodus and BBN would not have developped. So while those two companies figure out how to handle their relationship, maybe we could all learn from this and figure out a way to make peering work in a more scalable manner.
I'm all for this. What metric(s) shall we use? Traffic is an obvious one. Is that measured in packets per second, or bits per second? Since, in the USA (where the vast majority of traffic originates) circuits are provisioned as full duplex, _does it matter_ which direction the bits are flowing in? You _should_ have provisioned adequately for the flow, in any event. That is, if you care about how much bandwidth your customers can use. Assuming it does matter, in which direction does the value flow? Towards the recipient, because the infrastructure for provisioning hundreds of DS3s costs more, and the provider is not billing correctly for it? Or is it towards the content provider, because the infrastructure for provisioning hundreds of web servers costs more, and the provider is not billing correctly for it? Another one is route-miles. A provider with 1,000 route-miles of circuits 'obviously' has less value than a provider with 10,000 route-miles of circuits. How does speed of those circuits factor in? Perhaps the metric should be DS0-route-miles (64K-circuits per mile). Of course, one WDM dark fiber run of 250 miles would nuke this metric. How about dollar value of the network, in total bills paid to the telcos? Does this mean that networks that are owned by telcos have an almost-zero cost, or an "full retail price" accounting cost? Are networks that have reciprical agreements with telcos unduely penalized, or do they benefit? Another good metric might be number of customers. As a fine point, does an ISP transit customer of a network count as one customer, or does it count as 30,000 customers (the number of dialup customers serviced by that ISP). Does a web content provider count as one customer, or as 5,000 customers (the number of his hosting customers). What about special cases that offer free email accounts and/or web pages, that have thousands or millions of customers? How about number of network advertisments, or routes? Would this lead to silly announcements (de-aggregation) to 'equalize' the number of routing announcements? We haven't even got to the hard points, which is *how much* each of these metrics are 'worth'. We also haven't begun to address sites like gatekeeper.dec.com, ftp.cdrom.com, and the like. Nor networks with plently of suck bandwidth (but not much content) such as MSN and AOL. It's harder than it looks on the surface, folks. Clues appreciated. -- Dave Rand dlr@bungi.com http://www.bungi.com
On Fri, 21 Aug 1998, Dave Rand wrote:
Traffic is an obvious one. Is that measured in packets per second, or bits per second?
Bits, not packets. And I think that aggregate octets per hour or per day is a more reasonable way to measure it.
Since, in the USA (where the vast majority of traffic originates) circuits are provisioned as full duplex, _does it matter_ which direction the bits are flowing in?
Yes. In and out should be measured separately.
Assuming it does matter, in which direction does the value flow?
Here's the complex part. The value is not expressed in bits and it depends on the destination within the peer's network. I am assuming that we can map the IPv4 address space by city and that we can set some value to each intercity link. This means that a stream of bits entering a peers network in San Jose with a final destination in San Jose would be free. But if the stream of bits was destined for Santa Cruz there would be a small cost. And if it was destined for Sacramento there would be a somewhat larger cost because Sacramento is further.
Another one is route-miles. A provider with 1,000 route-miles of circuits 'obviously' has less value than a provider with 10,000 route-miles of circuits. How does speed of those circuits factor in? Perhaps the metric should be DS0-route-miles (64K-circuits per mile). Of course, one WDM dark fiber run of 250 miles would nuke this metric.
Route-miles between cities might be the metric for determining the cost multiplier. Of course, this would assume some standard city-to-city mileage and a standard boundary, rather like a LATA boundary, that would make Santa Clara considered to be equivalent to a San Jose destination since it is only a couple of miles from San Jose.
How about dollar value of the network, in total bills paid to the telcos? Does this mean that networks that are owned by telcos have an almost-zero cost, or an "full retail price" accounting cost?
I think that we need some standard way to calculate such costs. Since we are discussing how to account for regional transit I think that one way would be for the peer who would receive the payment to publish to a pricelist to their peers for city-to-city transit and use this pricelist. The arrangement would give the peers the option to buy circuits from the other peer at that rate. We would use some formula based upon how much traffic a reasonably standard intercity circuit could carry to determine the rate per bit over that link.
Are networks that have reciprical agreements with telcos unduely penalized, or do they benefit?
I would expect that they would neither benefit or be penalized.
Another good metric might be number of customers.
There is too much variation between customers for this to work.
How about number of network advertisments, or routes?
Same thing, too much variation, i.e. big aggregates and little ones.
We haven't even got to the hard points, which is *how much* each of these metrics are 'worth'.
Agreed. This is likely going to require an industry council to come up with the metrics and algorithms and specific numbers. Does this sound suspiciously like regulated peering? Yes. It is regulated peering but my intention is for the providers to work out the regulations within the industry and only have government involvement on a review basis. For instance such a council would need to satisfy the government that it was not acting in an antitrust fashion and they would likely review this on a regular basis.
We also haven't begun to address sites like gatekeeper.dec.com, ftp.cdrom.com, and the like. Nor networks with plently of suck bandwidth (but not much content) such as MSN and AOL.
I think that this suggestion addresses all sorts of asymmetry because I believe that the bottom line is that it costs more to transfer a given number of bits over a larger distance than a smaller one and that this cost can be quantified.
It's harder than it looks on the surface, folks. Clues appreciated.
I agree with this. However when I look ahead to a world in which the IP network is the only data communications infrastructure carrying voice, video, web, email, etc., then I think that the hard work must be done to create a scalable peering system. If we succeed at this then we will never have to go through the pain that the telephone network experienced with the forced AT&T government regulated monopoly followed by its forced dissolution still under government regulation. It is my opinion that the only way out from the specter of increased government regulation of the Internet is to pre-empt government action by working out a system that would be considered fair under existing antitrust laws. If we can go a significant way down the path to such a system, even if we have not yet implemented it, then I believe we will be able to secure support from the government for a self-regulated industry peering council. But that's just my opinion. I don't have all the answers. -- Michael Dillon - Internet & ISP Consulting Memra Communications Inc. - E-mail: michael@memra.com Check the website for my Internet World articles - http://www.memra.com
Assuming it does matter, in which direction does the value flow?
Here's the complex part. The value is not expressed in bits and it depends on the destination within the peer's network. I am assuming that we can map the IPv4 address space by city and that we can set some value to each intercity link. This means that a stream of bits entering a peers network in San Jose with a final destination in San Jose would be free. But if the stream of bits was destined for Santa Cruz there would be a small cost. And if it was destined for Sacramento there would be a somewhat larger cost because Sacramento is further.
Ok, here comes the Bell System all over again. This would have the effect of trickling down to the point that you would be charged more for downloading a web page from Germany or ftping an encryption program from Australia than you would for getting a web browser from Illinois. Start looking for peak/off-peak rates, different long distance billing, etc. Sounds like a good way to regress.
On Fri, 21 Aug 1998, Brian Pape wrote:
Ok, here comes the Bell System all over again. This would have the effect of trickling down to the point that you would be charged more for downloading a web page from Germany or ftping an encryption program from Australia than you would for getting a web browser from Illinois.
Start looking for peak/off-peak rates, different long distance billing, etc. Sounds like a good way to regress.
Anybody here remember the fairy tale about the goose which laid a golden egg? Who got egg on their face in that version? David Leonard
participants (4)
-
Brian Pape
-
dlr@bungi.com
-
M. David Leonard
-
Michael Dillon