0

I'm currently having trouble to explain the following behavior on an azure VM. The VM is a Standard DS4 v2 Promo (8 vcpus, 28 GB memory) with Ubuntu 18.04 freshly installed.

I generate some workload using stress --cpu 1.

Now if I check the Average Host CPU Percentage metrics through the Azure portal using 1 minute granularity I see an expected ~13% CPU Percentage whereas if I aggregate using max the CPU Percentage is ~97%.

It might be interesting to add that using stress --io 1 does not produce the same ratio between average and max aggregations (~7%, ~30%).

Also, I've been monitoring the guest OS and nothing else was running during the test.

Does anyone knows what cause such results ? Does Host CPU Percentage aggregate using max mean anything in that case ?

Regards.

Adrian Goudard
  • 13
  • 4

1 Answers1

0

Maximum of CPU over a time range is not very useful. There were some tasks maxing out CPU a long time ago (boot time, scheduled jobs), but that may not represent CPU use right now.

Graph data and look at the trend. Compare to guest load averages, which show runnable tasks over time.

Sustained high CPU, where it is at say 97% for several consecutive samples, is more interesting for a monitor or alert. Even then, depends on workload: 97% might be tolerable for a CPU-intensive background compile workload, but destroys response time on a busy web server.

--cpu vs --io is quite different work. --cpu does some maths keeping CPU resources busy, where --io does storage work in which the CPU mostly waits for disk.

John Mahowald
  • 32,050
  • 2
  • 19
  • 34