Data movement in high-performance computing systems accelerated by graphics processing units (GPUs) remains a challenging problem. Data communication in popular parallel programming models, such as the Message Passing Interface (MPI), is currently limited to the data stored in the CPU memory space. Auxiliary memory systems, such as GPU memory, are not integrated into such data movement standards, thus providing applications with no direct mechanism to perform end-to-end data movement. We introduce MPI-ACC, an integrated and extensible framework that allows end-to-end data movement in accelerator-based systems. MPI-ACC provides productivity and performance benefits by integrating support for auxiliary memory spaces into MPI. MPI-ACC supports data transfer among CUDA, OpenCL and CPU memory spaces and is extensible to other offload models as well. MPI-ACC's runtime system enables several key optimizations, including pipelining of data transfers, scalable memory management techniques, and balancing of communication based on accelerator and node architecture. MPI-ACC is designed to work concurrently with other GPU workloads with minimum contention. We describe how MPI-ACC can be used to design new communication-computation patterns in scientific applications from domains such as epidemiology simulation and seismology modeling, and we discuss the lessons learned. We present experimental results on a state-of-the-art cluster with hundreds of GPUs; and we compare the performance and productivity of MPI-ACC with MVAPICH, a popular CUDA-aware MPI solution. MPI-ACC encourages programmers to explore novel application-specific optimizations for improved overall cluster utilization.

(ABSTRACT) Today, heterogeneous computing has truly reshaped the way scientists think and approach high-performance computing (HPC). Hardware accelerators such as general-purpose graphics processing units (GPUs) and Intel Many Integrated Core (MIC) architecture continue to make inroads in accelerating large-scale scientific applications. These advancements, however, introduce new sets of challenges to the scientific community such as: selection of best processor for an application, effective performance optimization strategies, maintaining performance portability across architectures etc. In this thesis, we present our techniques and approach to address some of these significant issues. Firstly, we present a fully automated approach to project the relative performance of an OpenCL program over different GPUs. Performance projections can be made within a small amount of time, and the projection overhead stays relatively constant with the input data size. As a result, the technique can help runtime tools make dynamic decisions about which GPU would run faster for a given kernel. Usage cases of this technique include scheduling or migrating GPU workloads over a heterogeneous cluster with different types of GPUs. We then present our approach to accelerate a seismology modeling application that is based on the finite difference method (FDM), using MPI and CUDA over a hybrid CPU+GPU cluster. We describe the generic computational complexities involved in porting such applications to the GPUs and present our strategy of efficient performance optimization and characterization. We also show how performance modeling can be used to reason and drive the hardware-specific optimizations on the GPU. The performance evaluation of our approach delivers a maximum speedup of 23-fold with a single GPU and 33-fold with dual GPUs per node over the serial version of the application, which in turn results in a many-fold speedup when coupled with the MPI distribution of the computation across the cluster. We also study the efficacy of GPU-integrated MPI, with MPI-ACC as an example implementation, on a seismology modeling application and discuss the lessons learned. Dedication I dedicate this thesis to my mom, dad, and sister. iii Acknowledgments I would like to express my heartfelt gratitude to my advisor Dr. Wu-chun Feng, foremost, for his invaluable guidance and support throughout my M.S. studies. His belief in his students and encouragement to think independently to develop solutions to the research problems made the overall research experience greatly rewarding. His diligent work-ethic and dynamic personality had been a wonderful source of inspiration to me. It has been a pleasure …

Scientific computing applications are quickly adapting to leverage the massive parallelism of GPUs in large-scale clusters. However, the current hybrid programming models require application developers to explicitly manage the disjointed host and GPU memories, thus reducing both efficiency and productivity. Consequently, GPU-integrated MPI solutions, such as MPI-ACC and MVAPICH2-GPU, have been developed that provide unified programming interfaces and optimized implementations for end-to-end data communication among CPUs and GPUs. To date, however, there lacks an in-depth performance characterization of the new optimization spaces or the productivity impact of such GPU-integrated communication systems for scientific applications. In this paper, we study the efficacy of GPU-integrated MPI on scientific applications from domains such as epidemiology simulation and seismology modeling, and we discuss the lessons learned. We use MPI-ACC as an example implementation and demonstrate how the programmer can seamlessly choose between either the CPU or the GPU as the logical communication end point, depending on the application's computational requirements. MPI-ACC also encourages programmers to explore novel application-specific optimizations, such as internode CPU-GPU communication with concurrent CPU-GPU computations, which can improve the overall cluster utilization. Furthermore, MPI-ACC internally implements scalable memory management techniques, thereby decoupling the low-level memory optimizations from the applications and making them scalable and portable across several architectures. Experimental results from a state-of-the-art cluster with hundreds of GPUs show that the MPI-ACC--driven new application-specific optimizations can improve the performance of an epidemiology simulation by up to 61.6% and the performance of a seismology modeling application by up to 44%, when compared with traditional hybrid MPI+GPU implementations. We conclude that GPU-integrated MPI significantly enhances programmer productivity and has the potential to improve the performance and portability of scientific applications, thus making a significant step toward GPUs being 'first-class citizens' of hybrid CPU-GPU clusters.

We present a fully automated approach to project the relative performance of an OpenCL program over different GPUs. Performance projections can be made within a small amount of time, and the projection overhead stays relatively constant with the input data size. As a result, the technique can help runtime tools make dynamic decisions about which GPU would run faster for a given kernel. Usage cases of this technique include scheduling or migrating GPU workloads over a heterogeneous cluster with different types of GPUs.