An Online Self-Adaptive System Management Technique for Multi-Core Systems

This paper presents a light-weight online system adaptation technique for multi-core embedded systems running multiple applications. Thus far, online system adaptation techniques are restricted to reconfiguring resource management schemes such as operating frequency scaling or task-to-processor assignment. Additionally, in this paper, we enable to adapt the algorithm at runtime. That is, we selectively apply a suitable algorithm considering the system status when more than one algorithm candidates exist in the target application. We propose a generic and extensible self-adaptive framework with multiple applications in mind by providing generic programming interface that requires minimal changes in the legacy software code. It is shown that the proposed framework adaptively optimizes both resource management and algorithm selection with negligible performance overhead. The effectiveness of the proposed framework is experimentally proven with real-life examples.