Artificial bee colony-inspired run-time task management for many-core systems

Efficient resource and application management is one of the most complex and challenging tasks in high performance computing. Large-scale computing systems that contain hundreds, thousands or even millions of cores demand solutions that can operate in a distributed, robust, and scalable fashion. However, while hardware parallelism is relatively straight forward to achieve, this is not generally the case for software. This leads to under-utilization of the hardware parallelism as well as imbalanced load distribution causing inefficiency and hotspots. In response to this challenge, this paper introduces a novel distributed and decentralized run-time management algorithm. The proposed method is guided by an optimization model inspired by artificial bee colonies (ABC). While ABC have proven useful for optimizing large sets of numerical test functions, this is the first time they are applied in the context of many-core system management. The initial result shows that, the ABC model is promising in context of run-time management for many-core systems. It is also anticipated that the algorithms bio-inspired foundations will inherently enable scalability, reliability, and adaptation. We are showing initial experiments, where the initial results indicate the capability of our model to improve the thermal distribution across the system.