ThermaPlan: Improving Smartphone Thermal Management
Smartphone battery consumption remains a top concern of smartphone users and a critical constraint limiting the effectiveness of adoption of smartphone apps. Due to leakage current, device temperature has a large impact on processor and overall smartphone energy efficiency. The effect of temperature is exacerbated by the widespread deployment of multi-core smartphones. Mobile multi-core processors can begin to overheat quickly—within a few seconds—causing cores to be shut down and work to be lost. We are developing a new framework to manage device temperature at both short, medium, and long timescales, improving battery lifetime and increasing device energy efficiency by maintaining efficient operating temperatures.
Based on data collected from the PhoneLab smartphone testbed we have identified several core problems and opportunities in smartphone thermal management. First, poor interaction between core activation and CPU frequency selection decisions on Linux causes frequent thermal emergencies on Android devices. When one or more cores reaches a certain temperature, it must be shut down and cool before it can be restarted. Given the amount of cycles that are lost during the cooling period, it is more effective to preemptively lower the CPU frequency and avoid thermal emergencies altogether.
Second, the fact that the device acts as a heat sink creates an opportunity for medium-term thermal planning.Second, the fact that the device acts as a heat sink creates an opportunity for medium-term thermal planning. This is similar to the idea of computational sprinting (PDF) but at longer timescales—seconds rather than milliseconds—that create new challenges. Note that on modern multi-core smartphones most CPU configurations will cause overheating—frequently within a few seconds. So this form of computational "middle distance running" does not require running cores outside of their normal operating range or the presence of special materials providing thermal capacitance.
Finally, the fact that mobile smartphones move through a variety of thermal environments creates an opportunity for long-term thermal scheduling of delay-tolerant tasks. For example, app updates may be best to perform when the device is in a well-cooled environment rather than in a hot car or pocket. Realizing a service to intelligently manage long-term thermal scheduling requires addressing the fact that most smartphones do not have an ambient temperature sensor.