How do I calculate how big one fib/rib is in cisco BGP routingtable?
You can't.
The format of internal data structures in commercial routers is proprietary information. To know the size, you need to understand what kind of tree they use (radix trie, btree, tcam, whatever). And what kind of information they store for each route/prefix. No commercial company is going to tell you.
Different cisco products have different operating systems. They started with IOS for their classic line of routers. Then they built a totally new OS for their big internet backbone routers (cisco 12000, aka GSR, aka BFR/HFR). The BGP implementations on those 2 are completely different. Then there are the catalyst switches (no idea if they do BGP). But CatOS is also completely different. And then there is NX-OS on the cisco 7000 switches (which comes from a startup that cisco bought 10 years ago). Four different OSs. I bet no cisco engineer could answer your question even if they wanted to.
Having said that, there are commands to look at bgp memory consumption. Those might give you a rough idea. "show ip bpg summary" shows bgp memory consumption. But I bet not all of it. "show proc mem" might also be useful.
Also: cisco engineers might (and will) change and optimize code all the time. Or add features. Which can change the memory requirements. Probably not by big numbers, but they do change. With memory getting cheaper and cheaper, and cpu performance not increasing that fast, I would recommend any coder to use more memory when confronted with the classic time/space engineering question. (Do I store all info in memory ? That would increase memory requirements. Or do I let my code recompute things when needed ? That would increase cpu-requirements).
and are ipv6 PI adresses needed for multihoming? For some reason i thought they are but now im not sure.
The routing architectuture for IPv6 is precisely the same as it is for IPv4. Nothing has been done to solve the problem of multi-homing. So if you multi-home to two different providers, you can not use provider-assigned-adresses (PA). You need to get your own block of Provider-Independent (PI) addresses. You were right.
It's one of the reasons why routing people don't like IPv6. IPv6 was designed by applications people. No routing problem was solved by IPv6. All we got was wider addresses. For the rest, all old problems persist. It sucks.
