Abstract
Heterogeneous Chip Multiprocessors is a hotspot in CMP. In order to boost its potential power, the two-level scheduling architecture and the relevant DPK (Dynamic priority and 0–1Knapsack) algorithm is proposed in this paper to handle the scheduling problem for multiple DAG-structure hard real-time applications. Based on the characters of CMP, three dispatch queues in application level and task level are provided in the architecture to consider the multiple DAG-structure applications as a whole. In the application level, the DPK algorithm utilizes Laxity, a dynamic parameter to measure the current urgency of each application rather than the static deadline discussed in most other literatures. In the task level, the algorithm not only considers the sub-jobs of the application with the highest priority, but also at the end of each scheduling step, finds other proper unscheduled sub-jobs in any application to fill the idle time slice generated in this scheduling step just like the classical 0–1 Knapsack problem. According to the algorithm analysis and simulation experiments, with the DPK algorithm, the Successful Rate can be increased and the idle time of each processor is reduced.