On Thu 2006.09.28 at 22:30 +0200, Claudio Jeker wrote: > On Thu, Sep 28, 2006 at 10:45:24PM +0300, peter dunaskin wrote: > > Hello, > > > > We're a small ISP working on redesigning our network's topology and > > the overall architecture before replacing the existing > > infrastructure, which happens to be different variations of low cost > > router-boards running Linux. The ultimate goal thus being to run > > OpenBSD wherever we can. > > > > A job on top of our list would be to to migrate the core routers. > > Such solution would involve carp(4) and pfsync(4) -- hopefully making > > them fully redundant and a lot more solid network appliances than > > they currently are. > > > > But we're having a tough time figuring out how BGP should operate. > > In the existing scheme, an AS (autonomous system ) with 2 IP address > > blocks (193.x.x.0/24 and 194.x.x.0/23) is what holds our networks > > together. > > > > Our two optical lines, namely [upstream0] and [upstream1] are > > separated geographically (as illustrated in Fig. 1), themselves being > > interconnected by a wireless link. > > > > [upstream0] [upstream1] > > | | > > | | > > [core0]---------(wireless)---------[core1] > > |193.x.x.0/24 |194.x.x.0/23 > > | | > > [clients] [clients] > > > > Fig. 1 > > > > Question; Is it possible to make two BGP neighbors (one at core0 and > > second at core1) by using only one AS? We could get a second AS, but > > as we expect to have more networks around we would really like to set > > it up by only using one AS, not bothering ripe.net every time we > > expand the network. And yes, we plan to use OpenBGP. > > > > Yes, you can use one AS to announce two networks even if they are split. > You need to have a IBGP session between core0 and core1 and you should > configure core0 to announce 193.x.x.0/24 and core1 to announce > 194.x.x.0/23. The clou here is that if the wireless link goes down your > networks are still reliably reachable -- actually you can use any of the > links without major effects.
The issue if the wireless goes down will be that each network will not be able to "see" each other; that is, core0 will not have routes for 194.x.x.0/23 and core1 will not have routes for 194.x.x.0/23. This is because BGP will not accept route updates originating from its own AS. One solution is to use a static route on each core. > This should mostly do the trick. Perhaps you need to play around with > localpref, med or prepend-self to tune the network so that traffic for > 194.x.x.0/23 will prefer upstream1 over the wireless link but this mostly > depends on your policy how you would like your traffic to be routed. > > Use prepend-self to steer incomming traffic and local-pref for outgoing > traffic. > > -- > :wq Claudio
