DX NetOps

  • 1.  CPU monitor issue

    Posted Nov 15, 2017 04:03 AM

    Hi All,

    We have a Linux server on which CPU is continuously fluctuating like for every 5 seconds it reaches to 90% and then to 20% and our spectrum tool ends up with generating multiple alerts in a day. So, this server owner is asking us to generate the alert if only the CPU utilization of this server is above 85% continuously for 10 minutes. We tried to adjust the Duration Sec in Thresholds and Watches to 600 sec but no luck, alerts are still generating; Server Poll Interval (sec) is 300. Can some one please advise us on how to accomplish this requirement.



  • 2.  Re: CPU monitor issue
    Best Answer

    Posted Nov 15, 2017 06:20 AM

    There is probably something you did wrong, because that's exactly the purpose of adjusting the duration. Duration can only be a multiple of the polling frequency, but if your server is polled every 300 seconds and you put a CPU Monitoring configuration to a duration of 600, you should only get an alarm if the CPU is above the specified threshold for two consecutive pollings. There's one thing to consider however: Spectrum is doing real-time/instant monitoring: if the CPU is 0 all over the 10 minutes and 100 % used only for 5 seconds every 5 minutes and the metric is polled precisely during those 5 seconds, it will tell you that you're above the threshold (which is true in that case) but it's probably not what the server admin want.

     

    Instead of monitoring the CPU Usage for cases where the load is pretty inconsistent and unpredictable, you should rely on the CPU Load on 15 minutes for example. If the load (already aggregated at the server level) is above a limit that would be more accurate.



  • 3.  Re: CPU monitor issue

    Posted Nov 15, 2017 07:15 AM

    Hi Christophe,

     

    Thanks, now we have the same requirement on Physical Memory monitoring. The server owner wants to trigger the Physical Memory alert on a server, only when the Physical Memory usage of that server remains above 90% continuously for 10 minutes. can your please suggest how to accomplish this...



  • 4.  Re: CPU monitor issue

    Posted Nov 15, 2017 08:58 AM

    For the memory, you don't have any metric already computed on the device, so I suggest you do the standard threshold, putting 600 or 900 in the duration, knowing that memory is less supposed to rapid changes than CPU. Hope this will be sufficient!