Performance with Synchronization Workloads

Abstract

In the last two chapters, we used coarse-grain parallelism to evaluate performance of the hierarchical task queue organization. This chapter studies its performance under fine-grain parallelism. We use two workload models: lock accessing and barrier synchronization. The lock accessing workload models tasks with a critical section protected by a lock. In barrier synchronization, tasks iterate through a computation phase and a communication phase. Using these two workloads, we study the impact of spinning and blocking synchronization methods on the performance of the hierarchical and distributed organizations. The centralized organization is not considered here because it is not a scalable organization.