[Feat] Add cuda gpu busy threshold

[Wangmerlyn]

Add a busy threshold parameter to make gpu sleep while keeping.

[gemini-code-assist]

Summary of Changes

Hello @Wangmerlyn, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request introduces a significant enhancement to how GPU resources are managed, particularly during background 'keeping' operations. The core idea is to prevent unnecessary resource contention by making the GPU keeping process aware of the current GPU load, allowing it to intelligently pause its activity when the GPU is already busy. This should lead to more efficient resource utilization and improved system responsiveness.

Highlights

• GPU Utilization Management: I've implemented a new mechanism within the _keep_loop method of the CudaGpuController to monitor current GPU utilization. This allows the system to intelligently decide whether to perform the GPU keeping operation.
• Resource Optimization: The GPU keeping operation (_run_mat_batch) will now be conditionally executed. If the detected GPU utilization exceeds a newly introduced busy_threshold, the system will log a debug message and sleep longer, effectively pausing the keeping activity to avoid contention on an already busy GPU.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point in your pull request via creating an issue comment (i.e. comment on the pull request page) using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands.

Feature	Command	Description
Code Review	/gemini review	Performs a code review for the current pull request in its current state.
Pull Request Summary	/gemini summary	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in issue comments and review comments.
Help	/gemini help	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist is currently in preview and may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments to provide feedback.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Footnotes

1. Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

[gemini-code-assist]

Code Review

This pull request adds a busy_threshold to prevent the GPU keeper from running workloads when the GPU is already busy. The log message suggests a longer sleep when the GPU is busy, but the implementation uses the same sleep interval regardless of GPU state. I've provided a suggestion to align the code's behavior with the log message.

[gemini-code-assist]

The log message at line 177, "sleeping longer", is misleading because time.sleep(self.interval) on line 183 is executed regardless of GPU utilization. To align the behavior with the log message, implement a longer sleep duration when the GPU is busy.

Suggested change

	if gpu_utilization > self.busy_threshold:
	logger.debug(
	"rank %s: GPU busy (%d%%), sleeping longer",
	self.rank,
	gpu_utilization,
	)
	else:
	self._run_mat_batch(matrix)
	time.sleep(self.interval)
	if gpu_utilization > self.busy_threshold:
	logger.debug(
	"rank %s: GPU busy (%d%%), sleeping longer",
	self.rank,
	gpu_utilization,
	)
	time.sleep(self.interval * 2) # sleep longer when GPU is busy
	else:
	self._run_mat_batch(matrix)
	time.sleep(self.interval)