|
North American Network Operators Group Date Prev | Date Next | Date Index | Thread Index | Author Index | Historical Re: CIX routing service (was Re: AGIS/DIGEX)
> - CIX > - on-the-fly trust-the-providers routing updates > - media sensing through BGP peering sessions > - router could overload > simple peering at major exchange points. Instead of having people all > try to peer with each other, a lengthy process and complex to manage), > why not have CIX put a router in any exchange point and say, "Here CIX > members, peer with this router and you'll get routing to every other No. That wouldn't work w/ CIX. It'd have to be some other entity. Current CIX members wouldn't go for it. > CIX member at the exchange point." The "big ISPs" will still peer > with each other directly, even privately, but they have enough traffic > in between them to make it worthwhile. New or small ISPs would be the > primary benefactors. I've been thinking about this. Something like an MLPA, but without even the needed formality. Basically, a 2501 or 2 or 3 at exchange points acting as route reflectors might be handy for smaller ISPs who wanted to say "I'll peer with anyone here to get better connectivity to them". Since noone with > a few k routes is going to participate, 2501s will have more than enough memory, and a 2501 can handle 20-30 such peers without dying when reloaded. And all sessions would be screened against announcing 1673, 3561, 701, 174, 4200, 1239, 1, etc... (in no particular order) to the boxes, thus guaranteeing that screwups would be minimized. Eventually, perhaps an automated web site to allow people to build in the access filters that the boxes would apply to them - either on an as-path or per-route basis. Yes, this can be done simply (pretty simply, that is) through the RA, but the idea would be to sign up and say "Hey, I don't have time to deal with the peering requests, just peer with the MLPA-router and I'll hear you and you'll hear me". Someone would have to moderate/arbitrate it, but since noone would be getting transit through this thing, if someone was dropped from it for a day or two while they got their shit in gear nothing critical would be affected. Anyway, it's possible that the ISP/C might sponsor something - though the thought might be that you'd have to be an ISP/C member to participate but that noone would be required to peer or participate just because they were an ISP/C member. It's possible that we could cooperate with some other providers (all of whom have 24x7 NOCs) to pass of NOC-stewardship of the routers... > Some would say that CIX gets saturated. Yes, CIX-West was at one > time fatally saturated on incoming bit pipes (not necessarily CIX's > fault), but saturation can be reduced or eliminated by: > - having multiple CIX routers - one set at every major > exchange point (no, there's no bacbone in between). > - rate-limiting peers to a maximum of XX Mbps (ATM, yes; > possible on GigaSwitchs or FDDIs?). > - detecting then encouraging peers with more than X Mbps of > traffic (on average) to offload traffic to direct peering > or other exchange points. > - Using multiple iBGP routers at an exchange point. I think route reflecting is a much better idea than trying to build something which would actually pass data through it. > Then again, it doesn't have to be CIX; it just seems natural since > they've been running CIX-West for oh-so-long. A bunch of smart > BGP people can go out, get some money, buy the equipment, install it, > and have ISPs pay them for the service. > I don't think this is really a new idea, nor really my concern, but > I see two of my service providers struggling to get new peering and > think, "There's got to be a better way." > > ... just something to think about. It's a thought that I have had. I think this is doable and helpful. It seems that the larger exchanges are becoming multiple things to multiple people. Certainly ISPs in the DC and Bay areas are connecting to larger exchanges hoping to get access to the larger providers - but also (and more realistically) to get better connectivity between themselves. The key idea here (since all of this can be done through the RA anyway) is just to save the time of 30 providers trying to coordinate with 30 other providers re: "Do you want to peer with me?". Avi - - - - - - - - - - - - - - - - -
|