I have only 1 copy of the global routing table at the moment on an RB1100AHx2, and look at the CPU usage 
Tonight my second BGP peer is going to send me the full global routing table on top of what I currently have.
Is this to be expected on such a router, or should I buy a more powerful one?
Are you sure the CPU loads relate to BGP? A specific update process occurring? I wouldn’t imagine THAT many BGP updates in a day required for a full routing table. Try profiling the CPU and make sure it is the BGP process using all the CPU.
What are your traffic requirements? 100mb? 500mb?
Also what version ROS are you using?
At the moment we only have 50Mb available to us at this location, but are only using ~20Mb.
Using 6.10
I only assume its the BGP using this much cpu as we have never switched it on before and it only really seem to have occurred since we started using BGP.
Is there a way I can see what is driving the CPU so crazy?
/tools profile
It still amazes me the buttons ROS has that I hadn’t even found yet!!
That tells me that routing is about 44% and idle goes up to 90%
What is idle?
Idle is nothing - that is the percentage of CPU that is not doing anything at that point in time.
So its not specifically BGP causing the CPU load but possibly the route table lookups causing it. How big a table are you taking?
Around 850,000 routes, or 2 copies of the global routing table.
44% to Routing and 2 X full global Routing - is not that bad to a 1100.
I would probably consider a CCR to take two copies of a full table in BGP. If you can make it work and keep it stable, that’s great. However, as a general rule, we tend to put as much horsepower as possible on designs that involve MikroTik routers and full BGP global tables. CCRs are a great choice for this.
You might want to reconsider the above statement about CCR being a great choice for an Edge Router specifically for multiple BGP full table connections, at least until in ROS BGP process is able to use multiple cores.
Until then, an x86 (Multiple Core) router will continue to outperform the CCR.
Regards
Fair point, but the single core limitation is purely in software and we haven’t had much trouble taking in several full tables into the CCRs thus far. Add to that much more fowarding capability than an 1100, 10 Gig capability and the CCR is still a good choice for a peering router. Also, since we do many MPLS deployments, we tend to stay away from the x86 more and more due to MTU problems that have come up in VPLS transport.
I would rather live with a moderate limitation until the code is updated unless the design specifies a large number of BGP feeds.