A Multi-dimensional Scheduling Scheme for QoS-Aware Real-Time Applications on Heterogeneous Clusters

Multiple performance guarantees are increasingly needed in QoS-aware real-time applications. Unfortunately, most traditional scheduling algorithms only consider one or two dimensions of them. In this paper, we build a new scheduler model that is suitable for providing multiple services. Based on the scheduler mode, we propose a novel multi-dimensional scheduling scheme MDSS, which tasks timing constraints, QoS, throughput, fairness and load balancing into account. The MDSS includes three steps. Step 1 is used to provide real-time scheduling and the existing DASAP algorithm is employed in step 1. In step 2, we present a novel algorithm MQFQ that strives to enhance the QoS levels of accepted tasks and makes these tasks have fair QoS services, as well. Another new algorithm ITLB is proposed and used in step 3. The ITLB algorithms is capable of balancing load of nodes and improving throughput of the system. We have performed extensive simulation experiments to compare MDSS scheme with MDSR scheme, DASAP and DALAP algorithms. Experimental results indicate that MDSS improves QoS benefit, has better load balancing and has higher throughput over MDSR, DASAP and DALAP while guaranteeing high schedulability.