Self-awareness in heterogeneous, adaptive many-core architectures enabling proactive, self-optimizing systems

Meer over het boek

Adaptive, heterogeneous many-core architectures, like the Digital on-Demand Computing Organism (DodOrg), offer significant computing power and energy efficiency compared to homogeneous systems. However, their complexity creates challenges for developers and administrators in terms of management and utilization. This thesis aims to simplify this complexity by applying Organic Computing principles, focusing on self-x properties such as self-optimization and self-configuration. It proposes a holistic approach to achieve self-optimizing and proactive system behavior in these architectures, addressing key challenges. The first challenge involves establishing self-awareness within the observer component, which allows the system to assess and classify its current state. This requires a scalable monitoring infrastructure for continuous, coordinated observations capable of real-time data processing across diverse hardware. The second challenge is enabling self-optimizing behavior, where the system autonomously learns the best optimizations during runtime. Lastly, achieving proactive behavior allows the system to predict future states and initiate changes to optimize performance or prevent adverse conditions. The proposed solution features a flexible, hierarchical monitoring infrastructure for comprehensive system observation and data reduction. A lightweight, rule-based method enables self-awareness, while a Learning Classifier Syst