We have two SR5s in a RB532. One is providing backhaul and the other is connecting 7 wireless clients. Both SR5s are acting as APs using WDS. We are seeing about 70-85% CPU load even when nearly no traffic is going across the link! This RB532 is running 2.9.20 w/330MHz CPU frequency set.
The other end of the backhaul is a RB512 running at 330MHz and shows 30-40% CPU load. This is running NStreme on all routerboards.
The bridge interface on the RB532 is showing less than 10Kbps of traffic and less than 10p/s!
Why is the CPU utilization so high?
The link is still able to pass >20Mbps of traffic and we plan on moving the backhaul to a dedicated routerboard in the near future but we want to make sure everything is working well.
I can not confirm that … I have a P3 600 running 2 nstream 802.11a and 1 normal 802.11a just fine. No peaks at 100% CPU utilization with full blast traffic.
What i have noticed is that if you turn on nstream, depending on CPU you use, you constantly have a proportional % overhead on CPU utilization no matter what. On the other hand when you traffic test or you have the link going at full blast you will only see a bit of extra utilization (% again proportional to the CPU used).
What i can agree to is that RBs are not that fast but they perform well with 2 802.11a max 3 links (always thinking that you will never need the full blast of 802.11a) or a nstream for backbone and one or maybe more b AP interfaces. Always have in mind that some b cards utilize more the processor than others.
MT we need power on RBs. It whould be nice to have something around 1GHz, with a daughter board for more IFs. Nice powerfull and compact…
I would like to hear some comments from others… share your experiences
I use Nstreme on Routerboard 532’s. It works pretty good as long as you are not wanting to push more than 17-20 mbit. Anyway, as for high CPU usage; if nstreme is for a point-to-point link then turn polling off and set framer-policy to best-fit. It will significantly reduce your CPU usage.
Correct me if I’m wrong, but I’m pretty sure you only need polling if you are running point-to-multipoint.
Nice powerfull and compact… 
