Adaptive and Hierarchical Runtime Manager for Energy-Aware Thermal Management of Embedded Systems

Abstract

Modern embedded systems execute applications, which interacts with the operating system and hardware differently depending on type of workload. These cross-layer interactions result in wide variations of chipwide thermal profile. In this paper, a reinforcement learning-based run-time manager is proposed that guarantees application-specific performance requirements and controls the POSIX thread allocation and voltage/frequency scaling for energy-efficient thermal management. This controls three thermal aspectspeak temperature, average temperature and thermal cycling. Contrary to existing learning-based run-time approaches that optimize energy and temperature individually, the proposed run-time manager is the first approach to combine the two objectives, simultaneously addressing all three thermal aspects. However, determining thread allocation and core frequencies to optimize energy and temperature is an NP-hard problem. This leads to an exponential growth in the learning table (significant memory overhead) and a corresponding increase in the exploration time to learn the most appropriate thread allocation and core frequency for a particular application workload. To confine the learning space and to minimize the learning cost, the proposed run-time manager is implemented in a two-stage hierarchy: a heuristic-based thread allocation at a longer time interval to improve thermal cycling, followed by a learning-based hardware frequency selection at a much finer interval to improve average temperature, peak temperature and energy consumption. This enables finer control on temperature in an energy-efficient manner, while simultaneously addressing scalability, which is a crucial aspect for multi-/many-core embedded systems. The proposed hierarchical run-time manager is implemented for Linux running on nVidia's Tegra SoC, featuring four ARM Cortex-A15 cores. Experiments conducted with a range of embedded and cpu intensive applications demonstrate that the proposed run-time manager not only reduces energy consumption by an average 15% with respect to Linux, but also improves all the thermal aspects -average temperature by 14 • C, peak temperature by 16 • C and thermal cycling by 54%.

Learning-Based Run-Time Power and Energy Management of Multi/Many-Core Systems: Current and Future Trends

“…Statically optimized resource and power management are not likely to achieve the best performance when the input characteristics are changing. As a result reinforcement learning has been used for DPM [22][23][24][25][26], DVFS [18-21,52], or combination of DPM, DVFS and mapping [28,53,54] in embedded, desktop and datacenter domains. A detailed classification of existing RL based approaches for power/energy management is given Table 2.…”

“…Several principles have been followed for shutting the cores down, for example, greedy approach where a core enters into sleep mode as soon as processing on the core is finished and timeout approach that enters the core into sleep mode after certain time of idleness if no request is received within that time. Out of mapping, DVFS and DPM, they have been applied individually and in combinations as well, e.g., mapping in [15,16] and both mapping and DVFS in [27,28].…”

Voltage, Throughput, Power, Reliability, and Multicore Scaling

“…For instance, reducing may cause an increase in soft error rate (SER) [6]. Conversely, increasing causes an increase of temperature, accelerates aging and the probability of breakdowns [7]. This leads to min ≤ ≤ max ( 1 ) An execution may require more than a certain level of throughput to be meaningful [8].…”

Machine learning for run-time energy optimisation in many-core systems

“…The framework is based upon the Reinforcement Learning (RL) approach described in [11], [12]. The framework invokes workload prediction and appropriate V-F control to achieve energy minimisation for applications executed on a multi-core hardware platform.…”