Hello.
Today at evening we have seen an unusual cpu load on our main ccr1036 router. Usually it was 15-30% at peak, but today it was over 90% with no obvious reasons.
Going to tool>profile shows that all processing took 6% in total except for queues which consume almost 80% of cpu.
ROS version is 6.42.9.
Do someone see similar problems?
What is the cause of it? I’ve played with mangles (now marking connections before packets), reduced pcq queue sizes (from 50 to 10), changed queues types in the queue trees parents. I’ve noticed, that CPU load is lower with the RED queue in the parent queues of queue tree.
But even with these changes it spikes from 20% to over 60% of CPU load in peak times. I had no such problems on CCR 1009, there was a consistent 35-45% of CPU load on it.
This is really bothering me. Our customers complain about freezes in the peak times at evening.
What can I do to solve this?
Hey
Just a thought: as it is right now each child queue needs to borrow from parent for each and every packet. If limit-at (& max-limit) were defined, queues could work “internally” most of time and maybe reduce cpu usage?
Do keep in mind that sum of limit-at of children <= Max limit of parent
It’s hard to find any info about parent queue sizes. I’ll try to dig some more info.
Thanks for reply.
Hi, I don’t think playing with settings like that will help you very much. The problem is most likely caused by use of the “global” parent and the single tree structure for all queues. It is easy to max out a single core on a CCR with that traffic level with that many PCQs all connected to “global”. Moving to simple queues and flattening out this structure a bit in the process should resolve it.
Thanks for info!
Thanks
thanks
