0

I have an ubuntu that has 4 CPU cores, it is running an Nginx server, it has encountered around 10% intermittent packet loss recently. Please take a look at the dstat output as shown blow:


----system---- -------cpu0-usage--------------cpu1-usage--------------cpu2-usage--------------cpu3-usage------ ---system-->
     time     |usr sys idl wai hiq siq:usr sys idl wai hiq siq:usr sys idl wai hiq siq:usr sys idl wai hiq siq| int   csw >
20-06 18:00:33| 25  11  62   0   0   1: 23  14  46   0   0  17: 22  10  49   0   0  19: 22  10  68   0   0   0|  93k   41k>
20-06 18:00:34| 25  24  51   0   0   0: 21  47  15   0   0  17: 29  14  42   0   0  16: 24  18  58   0   0   0|  77k   31k>
20-06 18:00:35| 31  23  46   0   0   0: 27  40  14   0   0  18: 31  11  31   0   0  27: 32  15  53   0   0   0|  93k   39k>
20-06 18:00:36| 27  19  54   0   0   0: 26  44  18   0   0  12: 24   6  49   0   0  20: 27  13  60   0   0   0|  79k   33k>
20-06 18:00:37| 27  21  52   0   0   0: 24  42  18   0   0  16: 28   9  42   0   0  20: 19  19  63   0   0   0|  87k   36k>
20-06 18:00:38| 30  23  48   0   0   0: 24  44  18   0   0  15: 27  14  40   0   0  18: 23  24  54   0   0   0|  91k   38k>
20-06 18:00:39| 34  19  47   0   0   0: 24  32  20   0   0  24: 28  16  30   0   0  25: 30  16  54   0   0   0| 104k   50k>
20-06 18:00:40| 21  21  58   0   0   0: 19  19  30   0   0  32: 28  13  34   0   0  25: 18  20  62   0   0   0| 123k   59k>
20-06 18:00:41| 37  20  43   0   0   0: 29  34  16   1   0  20: 34  11  31   1   0  22: 28  24  48   1   0   0|  90k   37k>
20-06 18:00:42| 22  24  54   0   0   0: 22  36  18   0   0  25: 26  16  37   0   0  22: 21  25  54   0   0   0| 104k   45k>

Is the interrupt reaching up to 122k normal here? And if the amount of context switch is normal? Thanks.

Jepsenwan
  • 160
  • 3
  • 11

1 Answers1

1

The CPU time spend in soft-interrupts it quite high.

I would suggest you add the options -C all -i -a and check what device is giving you lots of interrupts. (Likely your NIC)

Dag Wieers
  • 346
  • 2
  • 7
  • Thanks Dag, do you mean checking the /proc/interrupts file to see what device is giving so many interrupts? I have employed the watch command checked it in a 1s interval, but is there any reference or standard to verify if the device is giving too many interrupts that make the CPU even fails to handle? – Jepsenwan Jun 25 '17 at 02:26