Change the `cpu-usage` EventCounter from int to double

[Tomius]

In the following line:

runtime/src/libraries/System.Private.CoreLib/src/System/Diagnostics/Tracing/RuntimeEventSource.cs

Line 78 in 235915e

_cpuTimeCounter ??= new PollingCounter("cpu-usage", this, () => RuntimeEventSourceHelper.GetCpuUsage()) { DisplayName = "CPU Usage", DisplayUnits = "%" };

The RuntimeEventSourceHelper.GetCpuUsage() returns an integer percent of all system resources used.

This means that on a machine with 128 cores, the smallest detectable change in CPU usage will be 128% of a single core, so a single threaded busy-loop application will report to be using 0% CPU according to this counter, when it's actually burning an entire core. This is very misleading.

I couldn't figure out the best area label to add to this issue. If you have write-permissions please help me learn by adding exactly one area label.

[Tomius]

Suggested label: area-System.Diagnostics.Tracing

Tagging subscribers to this area: @tarekgh, @tommcdon, @pjanotti, @safern
See info in area-owners.md if you want to be subscribed.

Issue Details

---

In the following line:

runtime/src/libraries/System.Private.CoreLib/src/System/Diagnostics/Tracing/RuntimeEventSource.cs

Line 78 in 235915e

_cpuTimeCounter ??= new PollingCounter("cpu-usage", this, () => RuntimeEventSourceHelper.GetCpuUsage()) { DisplayName = "CPU Usage", DisplayUnits = "%" };

The RuntimeEventSourceHelper.GetCpuUsage() returns an integer percent of all system resources used.

This means that on a machine with 128 cores, the smallest detectable change in CPU usage will be 128% of a single core, so a single threaded busy-loop application will report to be using 0% CPU according to this counter, when it's actually burning an entire core. This is very misleading.

Author:	Tomius
Assignees:	-
Labels:	area-System.Diagnostics.Tracing, untriaged
Milestone:	-

[safern]

I think this makes sense to me.

[josalem]

@mikelle-rogers, this is a good issue for you to pick up. Let me know if you've got time for this.

As for the change itself, I agree that this makes sense. While not a public API, this is a documented value, so the docs will need to be updated to reflect the change and when it happened.

One other thing that should be decided is whether this value should be out of N * 100 or just 100, e.g., 1200% vs 100% for a 12-core machine using every core. Currently, Linus/Unix code gets the former, but scales it to the latter to match the Windows variant.

[mikelle-rogers]

@mikelle-rogers, this is a good issue for you to pick up. Let me know if you've got time for this.

@josalem I am currently working on it. Who would you suggest I include in the discussion to decide what the value should be?