Hello, im running an RB1100AHX4 with about 200 simple queues for 200 PPPoE Tunnels.
Other than NAT for the PPPoE tunnels and a VPLS tunnel going to a border router, there is really not much running on it.
The problem im having is that the CPU is really uneven when dealing with about 150 simple queues and no more than 300mbps of traffic as you can see on this picture:
What could be the reason for most cpu cores being literally unused while core0 is apparently doing all the work ?
// Diego
I have observed the same behavior for mpls traffic streams.
Hello, just reporting back after a LOT of trial and error.
It seems that for whatever reason, at least on ARM (I have only tested this on 1100AHX4 and RB4011), the NATting is single threaded.
Removing nat from the router and doing it on a different box fixed all the issues.
Here you can see an RB4011 before removing the nat rules from the router.
And here, the SAME router just without nat in place.
I can reproduce this every time. So im not really sure now if this is a bug or if this is by design…
Any clues ?
// Diego
I am currently experiencing the same problem. 2 links are established through PPPoE on the ISP’s end. The core (RB4011) is currently configured with NAT, Load Balance, Simple Queues, and PPPoE Server. CPU0 is being used most of the time while leaving 1,2, and 3 as idle.
What if you spread traffic over several NAT lines?
Use mark-connection and PCC destination address and port as divider in Mangle.
You could do here with four NAT lines to split traffic. They NAT to the same gateway out or from the same gateway in.