Propagating thermal constraints to the scheduler

Session information has not yet been published for this event.

*

One Line Summary

Discussion on maximising performance within the thermal budget

Abstract

With the development of the power allocator thermal governor, Linux has learned to control SoC power consumption to control temperature. The governor uses models to translate device performance and usage to power consumption and vice versa. As part of the power allocator development, cpufreq cooling device now has the ability to build simple power models for the CPUs using OPPs.

In a SoC, different cores (or groups of cores) can belong to different DVFS domains with independent frequency control. Similarly, any restriction of capacities initiated by the thermal sub-system is also individually applied to the frequency domains.

The scheduler is responsible for assigning tasks to cores. It needs to be made aware of the capacity constraints to provide the available capacity to the highest priority tasks as well as maximise overall system performance.

In this talks, I’d like to discuss a prototype we’ve been working on to propagate the capacity constraints to the scheduler.

This work is based on top of the usage of energy models in the energy aware scheduler patch-set as well as the integration of the scheduler and DVFS.

Tags

thermal, performance, system, scheduler, dvfs

Presentation Materials

slides

Speaker

  • Biography

    Linux kernel developer working on power and thermal management.