"Suspend when non-BOINC CPU usage is above" -> Allow a grace period before triggering

[martixy]

Describe the problem

BOINC tasks are already low-prio by default, which means the processor can schedule around them on milisecond timescales.

If running the tasks has a noticeable effect on the computer, constantly starting and stopping tasks due to transient spikes in CPU usage can be very jarring. And is likely detrimental for the tasks themselves, as saved progress probably isn't perfectly continuous.

Fans spinning up and down, perfomance impacts (in my case performance improves while boinc is running, don't ask).

Describe the solution you'd like

Allow a setting to control how long the CPU must remain above the treshold before tasks are suspended.

Having some sort of hysteresis is the basics of control mechanisms.

Additional context

No response

[Grant-SSSF]

A lot of work for little benefit IMHO.
If people feel the need to suspend when non-BOINC CPU usage exceeds a certain level, they'd be better served by choosing a reasonable value- eg 85% (or higher).
Ideally the default value should be 85% or even 90%.

Single core/dual threaded systems are very much in the past.

[martixy]

See, the trouble is, even with 80-90% I often get undetectable transient spikes which trigger BOINC to suspend.

You know how task manager has a graph of CPU usage. It doesn't reflect the spike that triggers BOINC.

It delays resume after suspending, even if we're at 0% CPU. All I'm asking for is equal treatment. 😁

And why would it be a lot of work? It already has a mechanism to delay on one end, just put the same on the other end!

[CharlieFenton]

I agree this feature does not work week.

I also see frequent temporary "suspended CPU is busy" when the CPU is not particularly busy (on my Mac.)

I wonder if it might work better if we take several samples of non_BOINC CPU usage over a short period of time and suspend only if it stays above the specified threshold throughout that time period. I don't remember how often we sample it, but I suspect once per second, so perhaps suspend only if it stays high every sample for 5 seconds.

[davidpanderson]

A major goal of the BOINC client is to not affect user-visible system performance.
Hence the asymmetry: if it looks like the user is doing something CPU-intensive,
we want to stop computing immediately,
and we don't want to resume until we're sure that it's over.

If the way we calculate non-BOINC CPU usage is incorrect,
that's a bug and we should fix it.

[davidpanderson]

BTW we sample CPU usage every 10 seconds; we can't wait for a 2nd sample

[Grant-SSSF]

The sample rate could be increased to every 5 seconds. Of course this would make it more likely to detect higher CPU usage and therefore more likely to stop BOINC processing.
However, it would also mean BOINC processing would resume much sooner.

As far as fans ramping up and down is concerned, that can be sorted (depending on how basic your motherboard is), by the appropriate BIOS settings (on my older mid-range Gigabyte boards, it's labelled "Temperature interval." The larger the value, the longer the delay between a change in CPU temperature and the corresponding change in fan speed. Mine is set to 3).
The other option is using Fan Control or similar software to control the fans.

[martixy]

OH! Discussion!

A major goal of the BOINC client is to not affect user-visible system performance. Hence the asymmetry: if it looks like the user is doing something CPU-intensive, we want to stop computing immediately, and we don't want to resume until we're sure that it's over.

If the way we calculate non-BOINC CPU usage is incorrect, that's a bug and we should fix it.

This places me in an incredibly ironic situation.

I am that user now.

System responsiveness drops precipitously on my current setup when BOINC is NOT running. (More generally when computer has been running for a while and CPU usage is below a certain threshold. It's some kind of dumb windows bug that I have no explanation for.)

Also, is "user-visible system performance" the only permissible / valid criteria?
It is currently summer where I am. Fan noise varies very audibly with changes in CPU usage. It can be quite annoying.
(Although now that I think about it, there is the option to add hysteresis to the fan response, rather than the CPU usage. And then I also read Grant's comment, which suggested the same...)

Additionally, if made configurable, I believe a large part of effect on system performance would be absorbed by the OS's process scheduling - since computing tasks run at a lower priority.

At least on Windows, according to this answer:

If two threads want to use a CPU, and they're of the same priority, then they are scheduled via what is called "time-slicing" and over time, each gets about 50% of the CPU's time. Whereas if they're of different priorities then the higher-priority thread gets 100% and the lower- gets nothing.

I know BOINC is multiplatform, but I don't know how linux scheduling works.

---

BTW we sample CPU usage every 10 seconds; we can't wait for a 2nd sample

Like Grant said, we could increase the sample rate to 5 seconds. Then simply if sample 1 is above threshold, wait for sample 2, and only if it is also above the threshold, stop computing.

For the UI, that would mean a simple checkbox (since sample rate seems hardcoded).

I'm curious how the sampling works.

It's definitely not an AVERAGE over the sample period, since that would mitigate the effects of a transient spike, which contradicts the observed behavior.
Is it "if any 1-second average" in the last 10 seconds exceeds threshold, computing is stopped?
Or does the stop event not use the 10-second sample? (You said "we want to stop computing immediately".)

I'd also like to confirm something:
How are computing tasks made pausable/resumable?
(Is it implementation specific? Or does BOINC offer some special mechanism, like saving the exact memory state/processor context - like how OS stand-by works.)

---

I am developer myself (a web developer, but still not averse to the ways of the code).
Spoken from ignorance, it seems like an easy feature to implement for someone with no knowledge of the codebase - apart from the UI. I have no idea how desktop app UI works.

Feel free to educate me and offer advice/insight.

[AenBleidd]

How are computing tasks made pausable/resumable?

It is an mechanism every OS has.
I will not point you to the exact implementation for Linux or Windows in BOINC, but it's very similar to this: https://stackoverflow.com/questions/27844970/how-to-pause-resume-a-process-in-linux

[Grant-SSSF]

System responsiveness drops precipitously on my current setup when BOINC is NOT running. (More generally when computer has been running for a while and CPU usage is below a certain threshold. It's some kind of dumb windows bug that I have no explanation for.)

Long story short- if you are running an Intel system with a CPU that employs P/E (Power/Efficiency) cores on Windows you need a clean WIn11 installation with current chipset drivers & all current microcode updates to have the best chance for process scheduling & CPU power states to work properly (Win10 has had scheduling updates to support P/E cores, but apparently there are times where it requires a clean installation of Win11 for things to actually work as they should).

[martixy]

Long story short- if you are running an Intel system with a CPU that employs P/E (Power/Efficiency) cores on Windows you need a clean WIn11 installation with current chipset drivers & all current microcode updates to have the best chance for process scheduling & CPU power states to work properly (Win10 has had scheduling updates to support P/E cores, but apparently there are times where it requires a clean installation of Win11 for things to actually work as they should).

Oh, no. This predates hetero cores by a long time. My computer is very old now. Old enough to not support Win11 (officially).