Understanding FIRECRACKER LOAD AVERAGE

1

I wanted to get a better understanding of when we need to scale our application and to track performance, but I’m not quite sure how to read/test. I assumed the 0 - 1 on the Y axios was 0% to 100% CPU usage, but I see spikes going over 1 to 1.5, so I’m not super sure how to interrupt a lot of the charts on the metrics tabs.

2

Maybe need someone from Fly to correct this, but my understanding is that number is processes relative to CPU. So the number can exceed 1, even with 1 CPU.

e.g found this random explanation of general load average on Linux which I assume uses the same idea as Firecracker would.

Load average: 1.00, 0.40, 3.35

On a single core system this would mean:
  • The CPU was fully (100%) utilized on average; 1 processes was running on the CPU (1.00) over the last 1 minute.
  • The CPU was idle by 60% on average; no processes were waiting for CPU time (0.40) over the last 5 minutes.
  • The CPU was overloaded by 235% on average; 2.35 processes were waiting for CPU time (3.35) over the last 15 minutes.
On a dual-core system this would mean:
  • The one CPU was 100% idle on average, one CPU was being used; no processes were waiting for CPU time(1.00) over the last 1 minute.
  • The CPUs were idle by 160% on average; no processes were waiting for CPU time. (0.40) over the last 5 minutes.
  • The CPUs were overloaded by 135% on average; 1.35 processes were waiting for CPU time. (3.35) over the last 15 minutes.
3 Likes
3

Awesome, so if I’m running a dedicated 4 core CPU, and my average load is around 1, does that mean I’d be running at about 25% of max capacity?

4

This will need confirming since I’m not sure the general Linux load average matches the shown Firecracker number. But:

With a quad-core system, if you had a load average greater than 4.0 that would indicate all cores are at 100% capacity, and any overload will result in processes waiting for CPU time.

So 1 would be fine. Source:

2 Likes