Improving Scalability on Shared Memory Clusters