Evaluation of load balancing methods on embedded multicore systems and their effect on power consumption

With the increased availability of multicore processors in the consumer sector, chip vendors now start to build multicore processors intended for embedded systems. These processors are trimmed for cost efficiency and often omit parts like the cache coherency fabric. With the availability of those processors, the need for embedded operating systems which fully utilize their potential is growing. The two dimensional scheduling problem on such systems can be tackled in different ways. The focus of this thesis lies on multikernel operating systems and especially on their load balancing techniques. Three different algorithms were implemented as part of an existing OS. Different principles were chosen for those load balancing algorithms, one centralized, one decentralized, and one heuristic approach. Scheduling on each core was done in round robin fashion and was not subject of this thesis. To evaluate the algorithms a simulation module was designed and implemented to generate artificial load consisting of periodic tasks with implicit deadlines. To meet the rising energy constraints of such systems all three algorithms were extended to adjust the number of active cores according to the current computational demand.