Performance modeling and MPI evaluation using Westmere-based infiniband HPC cluster

In recent years, we have witnessed a growing interest in optimizing the high performance computing (HPC) solutions using advanced CPU and Interconnect technologies. These advances are driven by the fact that CPU manufacturers are facing extreme challenges in doubling the processors' speeds due to various reasons, such as the high temperatures and power consumptions. Therefore and as a continuation to Moore's law, recent trends in HPC systems have shown that future increases in performance can only be achieved through increases in system scale using a larger number of components, such as multi-cores and faster interconnects. In this paper, we evaluate a large-scale Infiniband cluster, equipped with Intel's latest Westmere processor using two MPI implementations. The paper presents the cluster configuration and evaluates its performance using various High Performance LINPACK (HPL) and Intel MPI (IMB) benchmarks. Our results show that system scalability can still be achieved with up to 87% efficiency when considering the right combination of MPI, interconnect and CPU technologies.