Scheduling hard-realtime parallel tasks onto the processor network with wrapped mesh topology

When scheduling realtime parallel tasks onto multiprocessor systems, it is very important to optimize the computing resource allocations to those tasks using load-distribution mechanisms. This paper proposes a static load-distribution mechanism, called the Double-Layered Load-Distribution (DLLD) mechanism, which schedules hard realtime parallel computing tasks. Due to the nature of the hard realtime computation, the deadlines of task executions must be previously analysed and perfectly satisfied. For this reason, static load-distribution schemes are preferred to dynamic ones. In order to optimize the scheduling results with respect to various types of parallel computing tasks and various architectures of parallel-processor hardware, the DLLD mechanism is comprised of two functional layers: the DF/IHS scheduler (the upper layer) and the DPSTS (the lower layer).