It's fucking up my design. The simplest design I can think of is get 3 ASRs, plug them all into a switch, and run iBGP between them all. But then I need to think about what happens to the traffic once it sees the BGP table. Normally? No worries, it just routes it to the best exit, just design pathways that are redundant with enough throughput. With Cisco ASRs? You burn licensed bandwidth routing within your AS. So now I need to make sure I don't do that, which means I'm adding another layer of routing behind the ASRs, which means those also need the BGP table, meaning I need an ASR there too for the same reasons I bought an ASR in the first place.
I know I can make it work with limited tables and additional routers that are cheaper (e.g. a cat 9300 running BGP). I don't need full tables, I want full tables. It makes routing more straightforward to understand. There's also a non-zero political impact. This is a rare instance where bragging about my badass triple homed BGP network that can handle 30Gig+ and takes full tables from everyone happens to be justifiable in business terms. Do you have full tables, multiple partial peerings, enterprise-class routers configured for optimal routing, and a class B to work with? I want people to be jealous, and I think I have the budget to do it.
When I close my eyes I imagine 3x 48 port 10gig SFP+ switches that can run full tables. I plug them all together in a full mesh, plug all sorts of stuff into the other ports, and it spits it out where it needs to go. I shouldn't need to license double throughput for "suboptimal" flows as routers route traffic within my AS.
Please think to include alternatives as you roast me, maybe I'll spot a gem I can work with. As far as I can tell a bunch of the Ubiquiti Infinity routers would do the job just fine and let me route everything how I'd like to, but they don't offer a line item for warm fuzzies, and I need at least 20 warm fuzzies.
No comments:
Post a Comment