In this talk, we will first introduce a novel dynamic load balancing scheme for AMR applications on homogeneous parallel systems (denoted as ParaDLB). It integrates a grid-splitting technique with direct grid movements, for which the objective is to reduce the parallel execution time. The potential benefits of ParaDLB are examined by incorporating it into the cosmology application ENZO code. Experiments show that by using this scheme, the parallel execution time can be reduced by up to 57% and the quality of load balancing can be improved by a factor of six, as compared to the original DLB scheme used in ENZO.
Further, our experiments show that simply moving a DLB scheme designed for homogeneous parallel systems to distributed systems will introduce significant overhead. Therefore, we propose a dynamic load balancing scheme for AMR applications on heterogeneous distributed systems (denoted as DistDLB). It takes into consideration: 1. heterogeneity of processors; 2. heterogeneity of networks; 3. shared nature of networks; 4. adpative feature of processors; and 5. adaptive characteristics of applications. The proposed DistDLB was implemented in ENZO code. Experiments show that it can significantly reduce the total execution time by 9-56% on different types of distributed environments, such as LAN-connected or WAN-connected or Grid environments.