RE: Exchanges that matter...
Good point on the CO issue. I was not suggesting that the telco CO was a direct analog for Internet exchange points. I might not have made my point clearly. Let me try again. What I meant was that the demand for traffic seems to be growing faster than technologies ability to deliver it. Even with the planned *new* technologies, I am predicting that there will be a point where the traffic of the exchange points will need to bifurcate simply to be able to process the load, i.e. more *smaller* exchange points all running at top capacity of the available technology. It's not exactly the same analysis of costs for the final mile that drove the telco CO deployment, but the idea is the same: sometimes available technology defines the strategy. Any better? des ---------- From: owner-nanog@merit.edu on behalf of Todd Graham Lewis Sent: Wednesday, December 04, 1996 3:46 AM To: nanog@merit.edu Subject: RE: Exchanges that matter... On Tue, 3 Dec 1996, Danny Stroud wrote:
Actually, I think the issue is not about moving. It is more about developing
*new* facilities now to handle the forecasted demands.
Certainly a valid pov.
As the Internet becomes more pervasive and the expectations of the users (and investors) become higher, we (the Internet access provider community) will need to have better, cheaper, faster, more resilient, etc. etc. networks.
Yep, uh huh.
I find it hard to fathom a completely pervasive network routing through a few exchange points.
Why not? More below.
As the national telco infrastructure evolved over the last few decades (with CO's on just about every corner) so will go the Internet.
But with most of the major backbone providers we're rapidly approaching POPs in every city as it is, and any finer granularity really doesn't make much sense, except maybe in the mega-cities. Exchange points are not analogous to COs; major routing problems ensue as the number of exchange points increase. E.g., if there were three EPs per continent, and if each major network connected to these three EPs, and a requirement for connecting to the EPs was that you have a fully redundant backbone, what would be the problem?
I admire the foresight of those attempting to develop new exchange points. I do not envy the uphill battle they have before them. des
I don't envy them either, but I'm beginning to question the "a chicken in every pot and a NAP on every corner" approach to network design. Of course, I don't strictly have to worry about these things; that's why I and AOL and most network operators have upstream network providers. __ Todd Graham Lewis Linux! Core Engineering Mindspring Enterprises tlewis@mindspring.com (800) 719 4664, x2804
On Wed, 4 Dec 1996, Danny Stroud wrote:
What I meant was that the demand for traffic seems to be growing faster than technologies ability to deliver it. Even with the planned *new* technologies, I am predicting that there will be a point where the traffic of the exchange points will need to bifurcate simply to be able to process the load, i.e. more *smaller* exchange points all running at top capacity of the available technology.
If the argument is merely that LAN bandwidth cannot handle the demands placed on them at the NAPs, I think that this is a weak argument for NAP proliferation. First of all, I doubt that NAPs do or will max out LAN technology. It may max out _cheap_ LAN technology, but so what? Gigaswitches are a hell of a lot cheaper than even ten T3s. If gigabit ether develops with anything like the speed that fast ether did, then doubly so. Plus, isn't ATM supposed to be infinitely scalable? Why don't you have OC-3 or even OC-12, whether it be ATM or IP/SONET? Assuming that LAN technology is the limiting factor (to which I can't really agree), why proliferate NAPs instead of constructing parallel physical networks at the existing ones? You can do bandwith aggregation at Layer 2 just fine if you try, and without increasing network complexity. (I know there are problems with this approach, but it can be done. Plus, fewer NAPs, more time to spend doing it right.) "But the routers can't process packets fast enough." NAP proliferation hurts rather than helps on this problem. Sure, you have more aggregate horsepower to throw at a problem, but from a simpleton's point of view it looks like a roughly exponentially more complex routing state from which to decide. In any event, it definitely doesn't help. You still haven't addressed the route flap problem. One argument I'm sort of suprised that you haven't made is that multiple NAPs make the network more redundant and, ergo, tolerant of failure. Of course, they don't. Three NAPs per continent is plenty to serve this purpose; anything over this is reckless.
It's not exactly the same analysis of costs for the final mile that drove the telco CO deployment, but the idea is the same: sometimes available technology defines the strategy.
Sure it does. I think that the argument that technology is forcing us to more NAPs is a losing one, although it's certainly theoretically plausible.
Any better?
Yep; keep 'em coming. __ Todd Graham Lewis Linux! Core Engineering Mindspring Enterprises tlewis@mindspring.com (800) 719 4664, x2804
On Wed, 4 Dec 1996, Todd Graham Lewis wrote:
Three NAPs per continent is plenty to serve this purpose; anything over this is reckless.
Woops. Three is a nice, round, theoretical number. Five is fine. Fifty is highly questionable to my mind. Thinking that more NAPs solves the problem is just flat wrong. Pardon me for my misstep here. __ Todd Graham Lewis Linux! Core Engineering Mindspring Enterprises tlewis@mindspring.com (800) 719 4664, x2804
On Wed, 4 Dec 1996, Todd Graham Lewis wrote:
Three NAPs per continent is plenty to serve this purpose; anything over this is reckless.
Woops. Three is a nice, round, theoretical number. Five is fine. Fifty is highly questionable to my mind. Thinking that more NAPs solves the problem is just flat wrong.
Hmmm.. Not sure "continent" is the right granularity here. In North America telecoms prices do not in general take enormous hikes when you cross state borders. In (say) Europe they do. Lines between European countries often cost more than lines between a given European country and the US. Also content and thus traffic is far more localised to each country due to language difficulties (well that part that doesn't go to English speaking countries anyway). So 3-5 NAPs (or whatever) per homogenous area (homogenous in content and in telecoms charging) perhaps. But ridiculous telco regulation within Europe and language differences makes a very strong case for at least one NAP per country (we dump about 50% of our UK traffic off in the UK). Alex Bligh Xara Networks
On Wed, 4 Dec 1996, Alex.Bligh wrote:
in content and in telecoms charging) perhaps. But ridiculous telco regulation within Europe and language differences makes a very strong case for at least one NAP per country (we dump about 50% of our UK traffic off in the UK).
What do you do with the other 50% of your UK traffic, Alex? -- Jim Dixon VBCnet GB Ltd http://www.vbc.net tel +44 117 929 1316 fax +44 117 927 2015
On Wed, 4 Dec 1996, Alex.Bligh wrote:
in content and in telecoms charging) perhaps. But ridiculous telco regulation within Europe and language differences makes a very strong case for at least one NAP per country (we dump about 50% of our UK traffic off in the UK).
What do you do with the other 50% of your UK traffic, Alex?
Country wise? Difficult to divide up traffic sourced in the UK from traffic sourced outside teh UK just at the mo, but we send the vast majority of the rest to the states. I'd guess 50% to LINX (UK), 40% to US one way or another and 10% to Stockholm SE. The figures are not too easy to get as we are transitting US networks back to Europe too. Alex Bligh Xara Networks
participants (4)
-
Alex.Bligh -
Danny Stroud -
Jim Dixon -
Todd Graham Lewis