TUPEC Algorithm Designing Based on Dynamic Task Scheduling

Almost all of the existing dynamic task scheduling algorithms adopt the strategy with firstly scheduling and then migrating. However, the task migration tends to produce heavier spending sometimes according to load balancing criteria. To solve this problem, according to application signal of task and value of the implementation cost for task, the paper proposes the TUPEC algorithm with establishing the relationship between the processor cores and tasks in mapping more accurately to avoid task migration. In order to make the processor load be balanced as far as possible, the TUPEC algorithm uses the task replication technology to short the empty waiting time of processor core. Aimed at the problem of task competition, TUPEC algorithm assigns the task to a processor core which has smaller length of local queue at the current time. The results show that the algorithm has much improvement in the performance of overall competing time, and the speed-up ratio.