Watching Out For System Overheating When Benchmarking On Linux

The newest module added to the Phoronix Test Suite is the start of a watchdog implementation with initially being focused on watching the system's thermal state to trip the benchmarking process if any of the selected sensors cross a defined threshold.

In working to prevent benchmarks from running if the system is reaching an overheated state where it could begin throttling, this watchdog was added in Git for the upcoming Phoronix Test Suite 6.4.

This module is automatically loaded when WATCHDOG_SENSOR= and WATCHDOG_SENSOR_THRESHOLD= are set as environment variables. E.g. WATCHDOG_SENSOR=cpu.temp,gpu.temp,sys.temp WATCHDOG_SENSOR_THRESHOLD=50 will interrupt the benchmarking process if the CPU, GPU, or system temperature exceed 50 degrees (C). You can adapt it to support any desired value and using any sensor supported by Phodevi (the Phoronix Device Interface) such as used by phoronix-test-suite system-sensors or phoronix-test-suite module-info system_monitor.

It's really simple and gets the job done for ensuring the benchmarking is halted if a sensor is tripped. When a trip occurs, the module will wait several minutes to see if the sensor crosses back into a normal value to resume benchmarking. If that doesn't happen, the benchmarking will stop.

There's already been a request to enable this watchdog by default in the Phoronix Test Suite, but for v6.4 will remain an opt-in feature. This also works as yet another nice example of the PTS modules with how our open-source benchmarking software can be further trivially expanded and will work with any benchmarks/tests complying with our OpenBenchmarking.org test profile specification.

Let me know if you have any questions or suggestions about this latest addition to the Phoronix Test Suite. If you or your organization is in need of any enterprise support, custom engineering, etc, feel free to contact us for commercial matters. Phoronix Test Suite 6.4 will be officially released in June.