Evaluation of the Effects of Thermal Interactions on Mobile Devices

In traditional computing platforms, the thermal problem emerged as the power density of internal components kept increasing. To address this problem, heat sinks started appearing, and in a few years, the limits of air cooling were reached. As the processor was the component that was the first one to reach these limits, the research in this field focused particularly on it, and a great number of dynamic thermal management techniques were proposed, with the goal of obtaining a good compromise between performance and thermal output reduction. However, in recent years, the ICT market has seen a progressive shift towards mobile devices. While some years ago the most used device for everyday computing was the personal computer, nowadays the focus is on smartphones and tablets, that satisfy the demand of mobility. These new platforms have various important differences compared to traditional ones like servers, desktops and laptops: as an example, they have a larger variety of sensors and communications interfaces, the integration of the components is very tight, the user touches and interacts more directly with the system, battery life is a very important aspect of the user experience and the device operates in a dynamic environment, where temperature can't be considered as a constant. These differences between mobile and traditional platforms are the factors that increased the relevance of the thermal problem. While it previously was a constraint on systems' design, it is now a factor directly harming the user experience. In fact, battery life directly depends on the device's power consumption, and there's a positive feedback loop between power consumption and temperature. Also, because of the tight integration, there is thermal interaction between parts of the device and consequently the feedback loop potentially concerns more than one single component, and this effect is exacerbated by the fact the presence of additional sensors and communication interfaces enable new types of workloads. Furthermore, as the user directly touches the system, high temperature can be uncomfortable. Because of this situation, mobile devices require a different approach to the problem. In first place, due to limitations on physical space, air or liquid cooling is not suitable for mobile platforms. As a consequence, software thermal management techniques have to deal with stricter constraints. Secondly, software techniques developed for traditional platform are not optimal, because they don't take into account thermal interactions, both between internal components and between the system and the external environment. Internal thermal interactions are, in fact, an aspect that needs to be investigated further. As an example, a DTM desktop technique that uses voltage and frequency scaling on a single component could be ineffective due to the effect of a high temperature, thermal coupled component. Thus, it is necessary to collect real temperature data from different components of the device. To achieve this, data is usually collected with sensors and counters that are placed directly in the system. However, this approach can't be used on mobile devices. Temperature sensors, that can be usually found on desktop components, are not always available or accessible on a mobile platform. In this work an infrared measurement setup has been used, to obtain thermal maps with good spatial resolution. The data resulting from running a set of benchmarks has been analyzed, highlight the thermal behavior of different components under different workloads. The other type of thermal interaction that has to be considered is the one between the device and the external environment. Mobile devices are exposed to a broad range of external environments, in which the external temperature, that was previously considered as a constant, has an impact on the system. In order to develop techniques to tackle the problem, it is necessary at first to understand and evaluate the magnitude of this impact on the performances of a device. Within this thesis work, this evaluation has been done running a set of benchmarks targeted at different parts of the systems at different temperatures, analyzing the results to show possible temperature related variations in the performances. The goal of this thesis is to evaluate the impact of thermal interactions on mobile devices, both between internal components and between the system and the external environment. This work covers a novel area of research and thus will serve as a basis for future work, especially in developing new dynamic thermal management techniques specifically designed for mobile devices.