On Sat, 29 Jun 2002, Richard A Steenbergen wrote:
Let me play devils advocate for a moment:
ooh danger ;)
On Sat, Jun 29, 2002 at 10:28:17AM +0100, Stephen J. Wilcox wrote:
I think this is the key point. Its common sense that peering with the downstreams will improve user quality of service by both reducing latency and taking unnecessary points of failure out of the network.
Think about it from the large tier 1's perspective. Lets say you are Joe Sixpack ISP, and they peer with you in one location. They now have to haul your traffic to and from this one location, wasting expensive bandwidth on their backbone. They also now have only one point where they send your
This assumes as per a previous point that they exchange routes outside the region. And as per your hot potato assumption even without your peering they will still be dragging your inbound from the point of interconnection nearest the source. And quit pro quo, assuming their big tier 1 peers do the same then it'll be the same on balance anyway (as they will carry the traffic in the opposite direction and losses/gains will cancel)
traffic, a single point of failure which can easily become congested.
Theres single points of failure whether with a peer or a transit if your network is of that size where you dont have redundant interconnects..
Making sure that it doesn't requires capacity planning, which again costs time and money. Heck even the router port and the cost of writing down your ASN in a central database probably costs them more than you are worth as a peer, even assuming that you pay the way to their doorstep.
Hmm okay this is valid, but really.. do they need to spend much time on you? Economy of scale and all that.. they can automate building filters, they dont need to worry about alarming small fry bgp sessions, once set up theres nothing much to do.
On the other hand, if they peer with your tier 1 transit provider, they probably have 6+ key locations on their network where they can send traffic, and plenty of capacity to support it.
Providing the point of interconnect that you have is big enough for you then this is not relevant
Why should they bother with you, what value do you add to them? Oh thats right, you think you should get access to their network for free. Not a convincing argument for them. :)
Hmm, I could take Paul Vixie's response about regulation - its their responsibility as big operators to ensure they dont exclude you merely because they're big and your small... (monopolistic) Steve
The point where I start to disagree with tough peering policies is the point where they turn exclusionary for no technical reason, for example an OC48 backbone or selling in 15 major markets. Why do you need an OC48 backbone? Maybe you engineered your network intelligently so you don't have to haul an OC48's worth of traffic around. All you NEED is the capacity to support the traffic being exchanged. And what does how many cities you sell in have to do with the exchange of traffic?
These policies exist for the sole purpose of excluding people who are not "one of them", even if they are otherwise technically capable of making a good peering partner. When all (or most) of the tier 1's get together like this to exclude potential new competition from having access to vital peering partners necessary to succeed, only bad government mojo can result.
Also don't forget that one of the quickest tests of tier 1 status is how much money you are blowing for no reason. After all, if you were REALLY one of them, you'd have purchased a billion dollar OC48 network "just because", and you would be doing your datacenter peering with SONET oc3's instead of ethernet "just because". If you're doing things cheaper, better, or faster, you're not really a "peer" of theirs are you. :)