4 Conclusions

In this paper we proposed a hierarchical partitioning and balancing strategy for the distributed implementations of SAMR applications. The HPA scheme takes advantage of hierarchical organization of the processor groups to restrict communications to smaller groups, thereby reducing the global communication and synchronization cost and exploiting concurrent communication. We presented two variant HPA scheme,viz. the Static HPA (SHPA) and the Adaptive HPA (AHPA). In the SHPA scheme, the total number of groups is defined a priori and the group topology is fixed or static during the execution of SAMR applications. In the AHPA scheme, the processor pool is adaptively partitioned into hierarchical groups at runtime to match the adaptive behavior of the SAMR applications. The HPA schemes are validated using experiments and simulations. It is experimentally shown that SHPA reduces communication costs as compared to the Non-HPA scheme, and reduces overall application execution time by up to $41\%$. AHPA dynamically partitions the processor pool into hierarchical groups that match the structure of the adaptive grid hierarchy. Initial evaluations of AHPA show that it can reduce communication costs by up to $70\%$. An experimental evaluation of the AHPA scheme is ongoing. Other variants of HPA are also quite promising - for example an Adaptive HPA taking into consideration the runtime system state. The meta-partitioner method as proposed in [5] can be incorporated into the HPA scheme framework to apply different HPA schemes for different system and application runtime characteristics.