The Role of Self-Awareness and Hierarchical Agents in Resource Management for Many-Core Systems

The future of Moore's Law is in jeopardy. The number of cores of many-core systems is steadily increasing for every technology node generation. Voltage scaling does not keep pace with the unabated decrease of transistor size. Higher leakage power and manufacturing variabilities are the consequences and lead to extremely critical power as well as thermal issues. These phenomena can downgrade the performance or endanger system's functionality as well as its reliability if they are not properly addressed. In near future, up to 90% of a many-core chip's area may have to remain inactive, this non-active area is termed Dark Silicon. These issues make the problem of resource management challenging. Future management systems need to be intelligent, anticipatory, and self-adaptive. They are supposed to integrate management of different aspects such as thermal, power, energy, performance, quality of service, process variability, occurrence of faults and aging effects, all in one. In this paper, we study the contributions in the literature focusing on techniques for dynamic resource management in multi-and many-core systems. We put emphasis on advanced approaches that exhibit learning, self-awareness, hierarchical monitoring and management. We categorize the existing approaches from a new perspective and argue that a self-aware hierarchical agent-based model is a proper methodology to monitor and management many-core systems, in particular when they need to deal with different competing goals. In addition, we evaluate the main objectives and trends in resource management of many-core systems in order to pave the way for designing future computer systems ranging from high-performance computers to embedded processors used in the era of Internet-of-Things.