On Wed, Mar 29, 2006 at 02:22:13PM +0100, tony sarendal wrote: > On 29/03/06, Claudio Jeker <[EMAIL PROTECTED]> wrote: > > > > On Wed, Mar 29, 2006 at 01:33:15PM +0100, tony sarendal wrote: > > > > > The second problem is, that I want to announce an external > > full-feed, > > > > > received with openbgpd, to my core-router. This works fine, but the > > > > > next-hop is the ip-adress of my external bgp-neighbor. But it should > > > > > be the ip-address of the border-router. > > > > > > > > of course! you should have an IBGP that makes your other routes have a > > > > route to that... like OpenOSPFD :) > > > > > > > > > or rewrite nexthop so you can run without an IGP. > > > > > > > ... and call for a lot of trouble when your network is growing. > > IMO this is a bad advice. 90% of all "set nexthop" usages are a hack > > around a network design error. No matter if it is on OpenBSD, cisco or > > whatever. > > > I don't agree. > A large network does not have to have an IGP to carry the BGP, > no matter size, it depends of the requirements of the network. >
I did not talk about redistributing BGP information into an IGP (that's totaly sick). I'm talking about the need for an IGP to glue your network correctly together so that BGP is working as it should. > A normal design is to use the IGP for to carry the network (links and > loopbacks) and punch the rest into BGP. A BGP-only design will allow you > to have features like per-link-direction routing for every single prefix > which wil allow you to do traffic engineering in a network per prefix by > just modifying origination point bgp communities. I.e I want this > specific prefix to have it's metric bumped by 100 on the LON->AMS and > 500 on OSL->STO to offload this traffic of away from those links for a > while. Or I don't want this prefix advertised over links which has MTU > less that 4470 to avoid fragmentation. > I think we're talking about the same thing you just used more words :) In large networks you use a "minimum" IGP (in ospfd you would only use interface statements but no redistribute rule) over that skeleton you run a BGP mesh. Without the IGP there is not skeleton and your BGP mesh falls totaly appart. In one point your right it is often better to redistribute (customer) networks via BGP. The filtering capabilities of OSPF are simply non-existent and so traffic engineering is often impossible. > You can do a lot of things in a bgp-only network which you can't do with > an IGP, as long as the limitations that bgp offers aren't a show stopper > for you, and as long as it suites the network you are running. > You can only run a BGP only network if your core network is consistent and to achieve that you normaly need some sort of IGP. > Also I have seen countless network problems just due to problems in one > of the many routing and forwarding layers, think protocol pancakes and > mpls, one protocol is simple. > Yep, I suffered too. Anyway, systems of that size should be administrated by people who know how routing works. Handicraft enthusiast do not get that far... -- :wq Claudio
