RB 3011 asymmetric cpu load

RouterOS General
1

Hi everyone.

I have an 3011UiAS, this guy is only doing ospf, is a “PE” router but, i can’t reach an ideal balance on the cpu load between the cores.
The load always are greater on core0 than core1. Someone have this problem too ? I’ve tried to balance the traffic and interfaces between the switches but without success.


srry for bad english :frowning:
cpu issue 3011.jpg

2

Can you post tools → profile so we can see what is making up the processor usage.

3

cpuprofile3011.png

4

“Asymmetric CPU load” should not be a problem at least until one of the CPUs is 100% busy. What real problem are you trying to solve? Does it cause you any real problem at all? Why do you think “symmetric load” is achievable?

5

Hi andriys. I’m not percepting any problem, just curiosity and the wish of make the routers more efficient as possible on network.
In other boxes we can achieve an cpu load balance by features, switch ports, whatever, is just curiosity and wish of make the thing more perfect as possible.

6

What a helpful post.

maybe go post on the ubnt forums instead ?

7

@Jajeblonsky is spamming the board left and right with this type of silly post in the last days

8

Ideal situation in packets path over router is to stick to one CPU core as long as possible ( until 100% load) , this is due the fact that transferring something between cores is very expensive (resource and time wise) in Linux Kernel.

Second, in these small multi-core CPUs manufactures are not very precise about reporting the load, in some cases only precise data is 100% load that is it, so i would loose sleep over this issue

9

Agree
Loadblancing multicore and new trouble come like out of order packet

10

Thanks macgaiver. As i said, the intention is just acquire a little bit more of knowledge and try to get the thing more efficient as possible. Thank you :slight_smile:

11

UPDATE:

I’ve scheduled the routers to reboot with 6.38.1 and enable mpls in the end of this dawn. Some minutes ago i finished to configure MPLS and, looks like MPLS table use much less cpu resources than default routing table. This is amazing.

Now I’ve got a lot of routers on the network but I have several routers that are doing ospf ( forwarding traffic to another routers ) and pppoe + queue in the same box, I will test mpls on these other routers to see if are possible to save some cpu load there. Thankyou all for the responses :slight_smile:
mpls.jpg