Forum Discussion

Noyan_137135's avatar
Noyan_137135
Icon for Nimbostratus rankNimbostratus
Mar 07, 2014

High CPU Guard

Hi all, I have a question about CPU mechanism of F5 devices. Is there any guard system or something for preventing very high usages of CPU caused by typically high connection requests? For example if the CPU of device rises to 90-95% with so many new connections received for one and half hour, does the CPU keep running high percentages of activity or cause a failover later?

 

Thank you for your answers.

 

3 Replies

  • Hamish's avatar
    Hamish
    Icon for Cirrocumulus rankCirrocumulus

    Nothing AFAIK...

     

    A quick check... Although the CFS (Completely fair Scheduler) is available from kernel 2.6.23 and supposed be the default nowadays... BigIP v11.5.0 uses 2.6.32... Quite old... I think 2.6.38 was out in 2010, but that doesn't take into account RedHats fiddling with various patches, they might have updated the CFQ to latest code).

     

    However, from a quick look into /sys/fs, it's not used to group any processes... (I'd expect to see a directory /sys/fs/cgroup IIRC if it were).

     

    H

     

  • So as you mention it seems the CPU behaviour will remain the same when receiving more connections than the average, and CPU usage percentage will run into 90-95% until the received connection amount gets back to normal. Right?

     

  • Hamish's avatar
    Hamish
    Icon for Cirrocumulus rankCirrocumulus

    TMM will use as much CPU as it requires...

     

    Whether that leaves enough CPU to schedule for the HA daemons to run or not... Well...

     

    ** FWIW I have seen a loaded BigIP failover when you search the logs from the GUI... But that was a 6400 running 9.something, & I'm not sure I've managed to starve current versions of enough CPU. (But then I don't work at that client any more and current clients don't utilise that much resource on their units...

     

    H