We are saving the configs of our TIKs periodically.
I found out now, that there is a massive CPU load when we are doing the “export” command on the router.
I tested it with ROS 6.32.4 and a RB750, which have very low CPU speed. With the CCR-Router and die RB1100AHx2 the problem is not there, because they are so fast. But on RB493G and RB2011 I have also 100% CPU for 2-3 seconds.
Is there a possibility to limit the CPU speed for doing the export? It would be much better that you have to wait 2 seconds longer, than bringing the CPU to so much load.
I think you should first prove that a process running at 100% CPU is actually disrupting service.
I have never seen that happen in Linux unless the process is at realtime priority… which the export most likely isn’t.
Remember it is a multitasking OS. The task of the kernel is to schedule the CPU between runnable processes
in a fair way. When one process uses (almost) 100% CPU it is an indication that the other processes have nothing
to do. Normally, the priority of a process that makes no event wait calls is automatically decreased so it cannot
take away CPU time from other processes.
Then we are doing the backup, pingtime of this deviced raises to aprox 150ms and we have also the same packet forwarding delay. So there is a service interuption.
Ok that is indeed not good. But the way to solve it would not be to reduce the CPU usage of this process
and thus lengthen the time, but to reduce the priority of the process that does this activity.
Then it will still show a lot of CPU usage but it will finish in the minimum time required.
It amazes me that this effect is visible. Normally the packet routing in the kernel has priority over the activities
of user processes like the program that does the export.