|
Sokar
4 posts
 |
I want to start off by saying that this is a pretty great site. I stumbled on it a little while ago. I wanted to get your take on using iBGP as an IGP instead of OSPF? If you have the people who know it and can manage it is there any reason not to use iBGP instead of OSPF? In an environment where you are doing multi-site each with multiple Internet links you have to run eBGP anyway. Why not just extend that and use it internally as well? It is a debate I have been having with colleagues lately and it seems to mostly end up as a religious debate where people tout what they know best.... Anyway thanks for any thoughts on it and keep up the good work of the community! |
|
netgyver
3 posts
 |
Hi, i think, it's not a good idea to run native iBGP as IGP. Best practisc is to run a link-state routing protocol (OSPF, IS-IS) as underlying IGP an run iBGP on top. Only this gives you a good convergence time in case of link-failures. BGP is distance vector and therefore converges not really fast. Hint: use OSPF or IS-IS for infrastructure addresses (loopbacks and core-links) and use iBGP for customer routes or non-core links. Regards, Alex |
|
ekaleido
4 posts
 |
Nevermind how you avoid a full mesh, because that can be fun in and of itself. You'll ask BGP for the next-hop for a subnet. Then you'll ask BGP for a next-hop for the next-hop. And so on and so forth. Read: use an IGP. If OSPF is too complicated, use something simpler. BGP is for reachability. |
|
Sokar
4 posts
|
@ekaleido - So you are referring to the processing overhead when you talk about the next-hop calculation? The full mesh problem only becomes a huge deal as the network grows in size. For smaller networks or where the connectivity is simple you can use route-reflectors can't you? Overall I agree that BGP shouldn't be used as an IGP for most situations. I have found that it works pretty well in smaller networks where there aren't a ton of hops. For Example a small two-site company with multiple internal VLANS using a switch to handle Inter-VLAN routing. Maybe a firewall off one VLAN connecting to a DMZ. In that kind of environment I have found that it can work. Maybe it isn't the best but with limited staff resources it allows them to run and learn only one routing protocol. It is an idea we have been tossing around but it sounds like it is not generally used like that.... |
|
jmbyrdwell
9 posts
|
BGP isn't really designed for your internal routing needs. You're gonna want a link-state protocol w/ hellos to ensure fastest convergence. Unless you have a a very large network you should really only need BGP when interfacing with your ISPs (if multihomed). If you do have a very large network and are worried about excessive talking from routing protocols you may want to consider breaking your network into Autonomous Systems, using and IGP within each, then connect them with BGP. |
|
ekaleido
4 posts
|
Sokar, yes what I'm referring to is the process for figuring out how to get around on a network that only runs iBGP. In a usual world, iBGP looks up a route, say 1.1.1.0/24 and is told to reach that network you go to the next-hop address of (again, for example) 3.3.3.3. But how do we get to 3.3.3.3 in your network? iBGP looks up a route for 3.3.3.3/32 and is told to reach that network through the next-hop of (again, for example's sake) 5.5.5.5. But how do we get to 5.5.5.5 in your network? This process repeats itself over and over again until the next-hop is a directly connected network, and this will occur on every hop in the path. In a small network, maybe this isn't a big deal. And if you intend to always be a small network, maybe the idea works for you. But it doesn't scale, you're creating extra work for your routers and the new complexity of what you're doing may end up not being easier for your staff after all. |
Viewing 1 - 6 of 6
- 1