This is correct, that in v5.0 you will see higher CPU usages with BGP. It is due to more frequent nexthop calculations which is made to address several problems where nexthops were not updated correctly. CPU usage depends on how frequently peer receives updates (more frequent updates will cause higher cpu usage).I've tested 5.0 on a RB1100 with a full BGP feed. CPU is at about 80-100% usage constantly. MT says it has to do with new routecalculation and that it's 'normal'. I'm going to run with three (maybe four) full feeds and will be buying x86 hardware for that.So is the BGP memory leak fixed?
I'm still running 4.13 that came installed on our RB1100, but it's leaking memory running BGP. We're taking full routes from one peer and just default route from another. So about 350k prefixes. Memory use seems to creep up... And now I'm wondering should I just go to 4.17 or 5.0 here.
(as you can see, we started running BGP mid-week 12)
IMHO, the RB1100 with 5.0 is not suitable for full BGP feeds.
We are working to improve performance, but until then users who experience too high CPU usage will require faster hardware.