Re: Policy Statement on Address Space Allocations
I think that we solved that problem by providing full information for all of the routes to routers within the area, and none outside. If you wish to correct me, feel free.
So everyone w/in the area can get to all of these sites (& since they have to carry full routes, they don't save any routing table space), and everyone outside of this area has no routes to any part of the area and thus can't get to any of these sites at all? Humm; if I understand you corretly, this seems to be a loose all around. --asp@uunet.uu.net (Andrew Partan)
I think that we solved that problem by providing full information for all of the routes to routers within the area, and none outside. If you wish to correct me, feel free.
So everyone w/in the area can get to all of these sites (& since they have to carry full routes, they don't save any routing table space), and everyone outside of this area has no routes to any part of the area and thus can't get to any of these sites at all?
Half correct. Everyone in the area carries full routes for the block. Everyone outside the area can listen to only the /8 advertisement. Once their packets get within the geographical area, then the routers will know what to do with it. So, no matter how bad a swamp this block becomes (and that would be avoided if possible, but...), it doesn't impact the outside world very much (1 advertisement and more-than-normal care on filter sets, since this is a special case). -george
Half correct. Everyone in the area carries full routes for the block. Everyone outside the area can listen to only the /8 advertisement.
So these providers are providing the free transit to their non-customers? This does not make any business sense; it will not happen. --asp@uunet.uu.net (Andrew Partan)
I still do not see why you insist this is providing free transit. If you're a backbone and connected into the area, within your network you advertise the /8 outside the area and the details within the area. If you're a backbone and not connected into the area, people can advertise the /8 to you and provide transit, or not, per existing peering and transit agreements. If you're a small provider within the area, you can talk to everyone else in the area and to the customers of backbones which come into the area. If you're a small provider outside the area, you probably buy transit to the area from your backbone along with transit to everywhere else. Everyone would probably want to special-case the block in router filters to make sure that they weren't being unintentionally used for transit by people who aren't supposed to use that route, but the upside is that those faraway routers carry one instead of hundreds or however many routes. -george
If you're a backbone and connected into the area, within your network you advertise the /8 outside the area and the details within the area.
This is the free transit case. Provider X is connected into the area, so provider X has all of the more specific routes for that area. Provider X is also connected outside of the area, so provider X has 3 choices. 1) Provider X can announce the aggregate outside of the area & thus give free transit to the whole area; or 2) Provider X can announce just provider X's customers outside of the area, thus defeating the gain from aggregation; or 3) Provider X can be paid by everyone else in the area to provide transit to the entire area to where ever else Provider X connects to. So either you can't aggregate (case 2) or you get paid by more customers (case 3) or you end up with free transit (case 1). Now lets look at the situation where there are many of these areas and you have some provider connected to more than one of them (as most big providers will be). For every area that provider Y is connected to, provider Y has to carry full routes for the entire area - no aggregation. Humm - I think that we just lost again. --asp@uunet.uu.net (Andrew Partan)
On Tue, 30 Jan 1996, Andrew Partan wrote:
Half correct. Everyone in the area carries full routes for the block. Everyone outside the area can listen to only the /8 advertisement.
So these providers are providing the free transit to their non-customers?
This does not make any business sense; it will not happen.
The proper solution is for all these companies to form a consortium. The consortium would run the NAP and contract with multiple NSP's for service. In that case, the NSP's are not providing transit to non-customers because the consortium is the customer and every ISP who joins the consortium gets multihoming reliability outside the region. Michael Dillon Voice: +1-604-546-8022 Memra Software Inc. Fax: +1-604-546-3049 http://www.memra.com E-mail: michael@memra.com
The proper solution is for all these companies to form a consortium. The consortium would run the NAP and contract with multiple NSP's for service. In that case, the NSP's are not providing transit to non-customers because the consortium is the customer and every ISP who joins the consortium gets multihoming reliability outside the region.
Ah - we may have something that works - we have someone (the consortium) being paid (by these companies) to provide (or further purchase) transit. However I'm not sure how this works of one (or more) of these companies decides to buy or provide transit on their own (outside of the consortium). Lets pick a company (call it X) that decides to do this. Now X's routes have to be known outside of the consortium's aggregate (since X is providing its own global transit and since X does not want to give free transit to the entire aggregate). Humm - this does not seem to scale. I suppose that if you find a set of companies that are all willing to be part of the consortium & just part of the consortium, then you could do addressing for this consortium as a whole. Hey! I think that we just invented provider (consortium) based addressing again. --asp@uunet.uu.net (Andrew Partan)
On Tue, 30 Jan 1996, Andrew Partan wrote:
I suppose that if you find a set of companies that are all willing to be part of the consortium & just part of the consortium, then you could do addressing for this consortium as a whole. Hey! I think that we just invented provider (consortium) based addressing again.
Depends on your definition of provider. I'm just saying that there appears to be room for another tier of provider in between the local ISP's and the national/global NSP's. NAP's and exchange points are popping up all over the place these days; it is only a small extension of the exchange point concept to a business that provides city-wide backbones (or meshes). It won't work everywhere, but it will work in some places and it will help control the growth of routes and that is good. Michael Dillon Voice: +1-604-546-8022 Memra Software Inc. Fax: +1-604-546-3049 http://www.memra.com E-mail: michael@memra.com
(Sending only to nanog, as per the recent coin-toss decision.)
From: Andrew Partan <asp@uunet.uu.net>
I suppose that if you find a set of companies that are all willing to be part of the consortium & just part of the consortium, then you could do addressing for this consortium as a whole. Hey! I think that we just invented provider (consortium) based addressing again.
One can also play tricks with tunnels, so that the folk with long prefixes inside the consortium's aggregate can get to the consortium without anything else needing to know a route for the long prefix. The packets travel farther (from source to consortium, and then from consortium to destination via the tunnel), but that's OK, because both the destination and the consortium are paying. --apb (Alan Barrett)
Michael,
So these providers are providing the free transit to their non-customers?
This does not make any business sense; it will not happen.
The proper solution is for all these companies to form a consortium. The consortium would run the NAP and contract with multiple NSP's for service. In that case, the NSP's are not providing transit to non-customers because the consortium is the customer and every ISP who joins the consortium gets multihoming reliability outside the region.
Forming a consortium is certainly an alternative for ISPs that aren't big enough, so that they don't provide a degree of addressing information aggregation sufficient to justify their routes to float throughout the "default-free" zone of the Internet. The consortium would acquire a (faily large) block of addresses, which would be partitioned among the members of the consortium. The consortium could run its own "backbone" (e.g., NAP) that would provide all members of the consortium with connectivity to other parts of the Internet. The basic connectivity (what I would call "route push") would be provided via this "backbone". The "backbone" would also perform address information aggregation into a single prefix. Additional connectivity can be acquired by the individual members of the consortium via "route pull", but this connectivity would be just to get better routers. This way a small ISP would have a choice of joining a consortium (and there may be a choice of consortiums to join) and get its addresses out of the consortium's block, or to connect to a large NSP and get its addresses out of the large NSP. However, one needs to understand the consortium model has its own issues ... Yakov.
(I'm reluctant to add to the noise, but what the hell - everyone else is partying :-) ) The space we are playing in is both a technical and an economic one, and the conversation stream to date veers widely around the technical issues and is completely off the planet economically speaking! Andrew half hit the issue with the comment relating to ISPs undertaking what is a variant of proxy aggregation to suppress routing table size when he noted that this is tantamount to "free transit". However this way he then inferred that free transit is impossible to sustain as the fatal flaw in this model, should be broken down further before dismissing it as completely flawed: a) forced proxy aggregation implies transit peering across ISPs Technically this (transit) is of course possible to construct. b) it will be economically infeasible IF we continue with this strange system of zero dollar interconnections we use as a peering model. i.e "free" is the problem here, not "transit". If you manage to provide a better model for interconnection which includes a rational economic model of interaction then, strangely enough, you then have a powerful tool you can use to address teh technical issue of scaling the routing domain. i.e. "free transit" is stupid, as Andrew indicates. "transit" is possible given a rational economic model of the transit interaction. In the same way that giving away IP addresses and giving away IP routing can only be described as a very bad case of irrational behaviour, especially when the underlying resource is under stress as it is at present, then I'd also note that giving away transit is similarly a case completely irrational behaviour! All this points to a desperate need for a more realistic economic structure to be used within a number of key aspects of Internet infrastructure. Thanks, Geoff Huston Andrew's comments:
Half correct. Everyone in the area carries full routes for the block. Everyone outside the area can listen to only the /8 advertisement.
So these providers are providing the free transit to their non-customers?
This does not make any business sense; it will not happen. --asp@uunet.uu.net (Andrew Partan)
participants (6)
-
Alan Barrett -
asp＠uunet.uu.net -
Geoff Huston -
George Herbert -
Michael Dillon -
Yakov Rekhter