CPU Usage Graph higher than actual - what gives?
Using small Linode for small game server development & experimentation hosting.
With the server running, my Linode CPU usage graph is pegged at 100% but according to all my system analysis utils my server daemon & support (via top) is only using about 7-10% of CPU cycles and the load doesn't even hit 0.20.
I don't want to be looking problematic at 100% constantly but it seems to me the graph is wrong… Unless I am missing something?
Please help, not sure what's going on or what needs to be done.
The Linode Manager graphs are best-effort, meaning they may not always be correct. This is because the graphing system cannot actually interact with your Linode since it has no access to your server. It would be worth installing Longview to confirm your resource usage. Since Longview is directly running on your Linode it would be able to give a more in-depth view.
The following two guides will be helpful if you aren't familiar with Longview:
Yes, the cpu usage graph is very…strange. Same OS and container app can show anywhere from 18-40% cpu usage at full app use. But whatever level it starts out at it seems to stay within the same range as long as it's still running. So if it starts at around 18% on the graph the graph will stay around there. If it starts at "30%" it stays about there. No idea why it's like that
At first I thought maybe it had to do with the shared core. Perhaps we are seeing estimated usage of the entire shared virtual core? And sometimes I just get put on a busy cpu? Or is it just showing account specific usage? If it's specific why would it be like that (fairly consistent, but only within the "levels" that vary)?
The CPU graphs are specific to your Linode and do not reflect estimated usage of the entire shared core. That said, landing on a busy host can result in CPU steal, which can affect your Linode's CPU usage. The graphs in the Cloud Manager are an estimate of your Linode's CPU usage, but it's always best to rely on the output of
top or to use Longview as dsmith suggests.