>"dre" wrote,
>""Howard C. Berkowitz"" wrote in message ... >> *sigh* you realize, I know, that we are talking about utterly >> essential concepts in working in the ISP space, yet these are things >> that don't show up on any Cisco test I'm aware of. This is part of >> why an R&S CCIE isn't much initial use to a large ISP without a good >> deal of OJT -- although they can probably deal with customer >> connections. > >Actually, they do show up on Cisco tests that I'm aware of, but I doubt >I can disclose the detail of that information, especially here. > >Also - I see it as the other way around... ISP's and whoever are still >going to take CCIE's over people with actual ISP "OJT" or even 10+ >years experience. Why? Because Cisco's a monopoly and just too >dominant. Worldcom and Level-3 will never have such success. I do remember a discussion at a NANOG meeting, however, where one of the senior people drew the analogy between (R) behind a name on a professional sports roster [rookie, for the non-USAians] and (CCIE). I'll grant that a CCIE can be useful quickly in doing things like customer BGP setup, once the Cisco idiocy about no static/default routes has been beaten out of them with an oversized clue stick. The kind of things I was considering, _some_ of which might be in C&S, but a lot of which I suspect would be considered more "design", could include: -- what grades of service do you offer? -- what communities do you set up for your customers and peers? what external communities do you accept? Do you use Geoff Huston's NOPEER? -- what's your strategy for customers multihomed to several of your sites but not to another ISP? RFC 2270? What's the role of NAT? -- to what extent do you run BGP (or more likely, what level of routes do you leak into) your maximum-bandwidth core, especially if it's (G)MPLS? Should the core know anything other than how to find next hops, with the intelligence of AS exits being in the distribution tier? For that matter, do you use best exit, closest exit, or a hybrid? What's your definition of "best"? -- if your local competition is selling based on lower hop count on their T-1's when you have three OC-3 hops, can you swallow your pride and common sense and tunnel so you show one hop? Or can you recognize your market position is such that you don't need to indulge in such idiocy? -- what's the longest NO-EXPORT prefix you will accept from customers? What will you export of other prefixes received, perhaps differentiating between customer and non-customer origin AS? -- How do you set up sanity checks? What should be the prefix limit? Rate limit on UPDATEs? Do you generate filters from a routing registry? Do you use strict or loose uRPF? Do you trust any peers? -- How do you deal with an attack where you want to get detailed traces, but you can't afford to do this in your core routers? UUNet presented an approach at NANOG to use a "shadow network" to which suspicious traffic can be transparently diverted, with detailed accounting, etc. > >> >the problem is that everybody pays Sprint (at least according to >> >Sprint), but the people who "might not pay Sprint, but shhh don't >> >tell anyone" (aka AOL) probably pay Worldcom. >> >> Well, probably a couple more. Think of who actually runs substantial >> national backbones of OC-48 or better. Sprint, AT&T, >> Worldcom/UUNET/whatever name du jour. There are some other major >> regional facility providers, including the Borgs...I mean, former >> Baby Bells. Now, it's perfectly reasonable that a Tier 1 could >> outsource the transmission, especially since they have the level of >> clue to write enforceable physical diversity requirements. > >The physical diversity requirements are enough for an average MBA >graduate to figure out how to finance.... dime a dozen. If you want to >and can sell it, you'll find a way to get into it (but not necessarily out). Fair enough, especially since the FCC came down hard on MCI for claiming to have sold diversity when it actually hadn't, even though there was no connectivity loss. Still, to what extent does Cisco teach the implications of automated grooming and the auditing of data layout record cards? > >Anybody can play with the big boys, especially if you have Warren >Buffett (Level-3) or Carl Icahn (XO) behind your company. These >people are also a dime a dozen in the high-end business world. The economics still aren't trivial. Bill Norton has probably explored this to the greatest extent -- what is the strategic business model of encouraging peering versus finding the last possible way to sell transit? Remember, peering also means that the big guy may learn routes he might not learn from someone who isn't directly BGP-connected to him. How important is this to customers? > >> >Speaking of AOL, >> >where do they fit? Tier 1, 2, or 3? They don't have enterprise IP >> >transit customers... they aren't really a transit-AS.... but they >> >are arguably larger than most ISP's that consider themselves >> >Tier-1's. Same with MSN. >> >> I don't know that much about their internals, so I'm making an >> educated guess here. My recollection is that they outsource >> substantial parts of their transit requirements. > >They outsource plenty of their transit requirements... but they >charge more than standard transit prices for their sub-1k routes. >Explain this. /begin singing from Cabaret "money makes the world go round the world go round the world go round /end singing > >> Frankly, I'd call them much closer to an ASP than an ISP. > >oh the blurry lines... this is very true ;> I very much agree >with you on this point. I once ran out of room on a 4x8 whiteboard showing the permutations of local exchange, intra-LATA toll, inter-LATA intrastate, interstate, competitive local loop/dial tone/switching, etc., under the rules of the California PUC, which probably changed while I was drawing. > >> If you want to go this route, than we really need to introduce >> categories beyond ISP. You have major access providers/broadband >> aggregators/dial wholesalers. You have application service providers, >> which blurs with content provider. For that matter, you have >> wholesale content delivery providers like Akamai. > >Again, agreed. Unfortunately, many of the people driving business >behind "the ISP indsutry" were previously using the "Voice" model >and selling like a traditional telco. Clearly, the retail industry and >especially energy/defense industries are way ahead of the telco >industry selling/mktg/general business strategies and models. So >it's another repeat of a sad story.... the ISP industry could end up >just like the telco industry... or worse, the automotive or consulting >industries. Yet there are things to be learned, even if not copied, from the telco industries. The mechanism of separations gives an economic model for end-to-end QoS without specific contracting. Geoff Huston's books are excellent on the various economic models here. Unfortunately, we still have a large body of enterprise-oriented people that don't understand why they shouldn't automatically get "optimal" (whatever that means) routes across the public network. > >> >The most important provider for most business today is not any of >> >the above types, surprisingly. It's the pen-ultimate Tier-1... >> >the Exchange Point Provider. There has to be someplace where all >> >these people meet to exchange traffic and connect with circuits... >> >and in today's world... it's the Exchange Point. >> >> Yes and no. Now, most details are proprietary, but my impression is >> that more pure bandwidth is exchanged between "high-level" provider >> through direct private peering connections (private lines or >> equivalents) than through exchange point fabrics. Confusing this > >And the cost of a 10/100 interface or two, or even sixty, is....? It's not a cost, but a price and competition issue. Sure, the cost of a copper or even fiber cross-connect between cables is trivial as to hardware, and not that great even if you load it (as you should) with operational expense. But the reality is that some exchange points do not allow private peerings, or charge a very large amount for them. > >> even further is the presence of private peerings both between the >> cages/racks in an exchange point (a piece of fiber), and the less >> visible private peerings that take place in the exchange point >> building but not in the exchange point proper (i.e., fiber >> connections down in the cable vault). > >Or cheap $20/month copper cross-connects. To everyone >else in the facility. Again depending on the exchange rules. It's not a given that you enter into a universal multilateral peering arrangement at a larger exchange, although this does seem to be the norm for the smaller exchanges. > >> There's a delicate economic balance whether bilateral/multilateral >> peering at exchanges gives you enough routes, and also enough >> economic fairness. Bill Norton has published rather extensively on >> this topic, and presented at multiple NANOGs. > >Well it really comes down to returns... and the WACC/discount/ >hurdle rates your company employs and how they measure when >the company goes into big capital or operating expenditures. > >Unless you run your own business and can do whatever you want. >There's always power in the normally powerless (Tier-5) and >powerlessness in the typically powerful (Tier-1). > >> The reality is that at some (there's no rule that applies to all) >> exchange points, one provider may still buy transit from another and >> peer with yet others. Different exchange points only permit peering >> without financial exchange. > >That's not true. There's no rules to who allows peering or not by >the exchanges... that's like saying the NYSE is controlled by MSDW. >Well, actually, you might have a point there... but for 99% of cases, >you can probably get away with enough to make returns viable >regardless of exchange politics. The _exchanges_ very much have rules, but if you mean there is no standard set of rules for exchanges, that's correct. I found it extremely interesting to participate in the RIPE/European exchange discussions, where the economic models are far more varied than in the US, and there are many more exchange players. Now, to appreciate this fully, you have to know some of the people, but I attended a RIPE exchange operators meeting where the CEO of a new exchange business briefed on their business model -- buy out all the local exchanges, or, if they wouldn't sell, set up a competitor that would kill them. The CEO was a banker by background, and, if she wasn't in a proper business suit, I'd have sworn she was the Borg Queen. As the audience went into super-flame mode, one of her advisory board members got up to try to calm things down (all sides were hysterical). If you can picture a situation where Randy Bush is being the calm diplomat... > >> One incentive to greater exchange point usage may be IPv6, the >> unicast address structure for which reserves top-level identifiers >> for exchange points as well as major transit providers. This would >> allow geographic rather than carrier-based aggregation. > >But there is a lot of security, homeland protection concerns >currently with geographical vs. carrier-based aggegation. Scale >free network topologies built on BGP have influenced both IPv4 >CIDR routing and IPv6 future routing. I'm not totally clear which way you are leaning on the security point -- it would seem geographic aggregation lends itself more to national control, given the larger carriers are multinationals. Also, I'm not clear on what you mean by scale free BGP. BGP is running into severe scaling problems, although they aren't necessarily obvious yet. I participated in a panel discussion of this at the Internet Society Stockholm meeting -- I think the slides are up at nexthop.com, under some of Sue Hares' presentations, but I'll also put them up elsewhere. A new routing scalability group just formed in the IRTF, but I wasn't able to attend the meeting -- the economy hits me as badly as anyone else ("Will architect networks for cat food and other valuable considerations"). Unfortunately, the usage patterns of the Internet have changed, basically with end user multihoming, so that the logical topology has flattened much more than provider-based aggregation and CIDR assumed. Routing table memory size isn't particularly the constraint any longer, but processor churn and overall latency of accurate topology information. I won't say there is anything approaching a consensus, other than BGP/path vector probably has theoretical limits. Whether map exchange, control theory, hydraulic, or other models replace it are a completely open issue. Message Posted at: http://www.groupstudy.com/form/read.php?f=7&i=71597&t=71508 -------------------------------------------------- FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]