Title: Designing HPC, Big Data, and Deep Learning Middleware for Exascale Systems: Challenges and Opportunities

Abstract: 
This talk will focus on challenges in designing HPC, Big Data, and Deep Learning middleware for Exascale systems with millions of processors and accelerators.  For the HPC domain, we will discuss about the challenges in designing runtime environments for MPI+X (PGAS - OpenSHMEM/UPC/CAF/UPC++, OpenMP, and CUDA) programming models by taking into account support for multi-/many-core, high-performance networks, GPGPUs, and Intel Xeon Phis. Energy-aware designs and co-design schemes for such environments will also be emphasized. Features and sample performance numbers from MVAPICH2 libraries will be presented.  For the Big Data domain, we will focus on high-performance and scalable designs of Spark, Hadoop (including HDFS, MapReduce, RPC, and HBase), and Memcached using native RDMA support for InfiniBand and RoCE. Designs to utilize HPC clusters with Parallel File Systems (such as Lustre) for Big Data applications will also be presented. We will present how high-performance virtualization schemes can be designed for HPC and Big Data with SR-IOV and Containers. For the Deep Learning domain, we will focus on co-designing popular Deep Learning frameworks (Caffe and CNTK) with MVAPICH2-GDR MPI library to get high-performance and scalability.