When developing complex systems like schedulers or resource managers, create simulation-based testing frameworks to validate functionality in scenarios that cannot be easily reproduced in development environments. This approach enables testing of hardware configurations you don't have access to (GPU/NPU nodes), large-scale cluster behaviors, and parameter...
When developing complex systems like schedulers or resource managers, create simulation-based testing frameworks to validate functionality in scenarios that cannot be easily reproduced in development environments. This approach enables testing of hardware configurations you don’t have access to (GPU/NPU nodes), large-scale cluster behaviors, and parameter changes before production deployment.
Simulation testing should provide visibility into system decision-making processes and support regression testing. For example, when testing a scheduler simulator, ensure it can show “how many nodes are there, which nodes are filtered for what reason, and how is each node scored” to enable thorough validation of scheduling logic.
Consider implementing simulation frameworks that:
This testing approach is particularly valuable for validating the impact of configuration changes, such as “after change the mostrequested.weight, if the average wait time of big task is shorter than before.”
Enter the URL of a public GitHub repository
