Hi Simon, On Thu, Oct 08, 1998 at 12:14:43AM +1300, Simon Blake wrote:
For the last couple of months I've had a linux box running the MRT route daemon as an experimental BGP route server, with links between a few organisations on Citylink who were happy to participate in a trial.
The main aim of the experiment was to consider ways that Citylink connected organisations with BGP capable routers but without AS numbers might exchange routing data, so that they can send data directly between each other, without having to go through one (or several) ISP routers.
Is it reasonable to assume that all networks reachable in this manner (a) will be numbered on globally-unique addresses, and (b) will have a WIX next hop within 202.7.0.0/23?
As a rudimentary route server, MRT seems to work really well (it lacks some of the RIPE tools of the Merit RA, but on the other hand, it compiles under Linux :-). We ended up setting up MRT to peer with the organisations who had AS numbers, and then setup all the other routers in AS number 65502 (chosen at random) as static routes within MRT, and redistributed them into BGP. Turn off synchronisation on the routers in AS65502, and everything seemed to work as advertised (this is speaking from memory, I haven't looked at it for about 6 weeks).
This is a cool idea, but I have a minor reservation based on the setup you described. I have a problem with the use of a "private use" ASN (rfc1930) for essentially the same reason that I object to the use of rfc1918 addresses on WIX - this is shared infrastructure, and we should use globally-unique numbering (for IP addresses and for ASNs). To not do so may/will cause problems - we should not have to worry about _our_ use of any private-use ASNs or IP addresses within _our_ network conflicting with anybody else's. The way to fix this is for the Citylink route servers to operate under a globally-unique ASN, and to essentially hide all the private-use ASNs to other networks when advertising the routes on to ISPs. Ciscos will do this automagically for private-use ASNs (64512 to 65535), and I assume mrt can do something similar? Since all the BGP speakers (with global or private-use ASNs) will learn routes with next hop addresses on the 202.7.0.0/23 network, they will route traffic to them directly (since this is a connected route). This scheme very closely follows that suggested in rfc2270, with Citylink taking the role of the "local provider", talking BGP to their local customers and providing transit to other "global providers". I'm pretty sure this is also how most other big exchanges function with route servers. I'm not sure how your redistributed static advertisements fit into the picture - surely only traffic in one direction would escape the ISP's router in this case? Or was that just part of the mrt testing?
Citylink is now at the stage where we've got customers pestering us to incorporate them into the system, and a management itchy keen to take some money off of these people. So I guess what I'm after from the list is some opinions on whether this is a good idea, and I'm on the right track. Obviously, for this to be more than a toy I need the ISP's on Citylink to participate in such an exchange, so recommendations on making it easier for them to be involved are welcome.
I think this is definitely a good idea, and will help everybody out
tremendously, as long as (a) the route servers operate under a globally-
unique ASN [as above], (b) everybody commits to filtering the
advertisements they receive, and (c) those filters are built from
consistent, published policy (e.g. in the IRR).
I might also add (d) as long as there are at least two route servers,
preferably located in different buildings in Wellington :)
Joe
--
Joe Abley