CSE 6429: Network-Based Computing for HPC, Cloud, Big Data and Deep Learning

Instructor: Prof. Dhabaleswar K. (DK) Panda
Autumn 2016

Course Number: 6429 - 0010

Class Number: 8371

Credits: 3

Course Time: TBD

Classroom: TBD

This research seminar is offered every semester.

Instructor: Prof. Dhabaleswar K. (DK) Panda

Objectives

During the last several years, networking and computing technologies are undergoing rapid growth and leading to low-cost, high-performance, and commodity computers and networking components (switches, routers, and adapters). Significant advances are also taking place for GPGPUs, Accelerators (Intel Xeon Phis), NVMe-SSDs. These advances are giving rise to a new computing paradigm known as Network-Based Computing where computers distributed over LAN and WAN can be used together to provide computing environments for a wide variety of application domains (such as scientific computing, enterprise computing, cloud computing, Big Data and Deep Learning). Such systems are typically connected over high performance interconnections such as InfiniBand, Omni-Path, 10/40/100 GigE Ethernet, and RoCE.

Several challenging research issues have to be addressed in designing such network-based computing systems. These include issues related to interprocessor communication, collective communication, synchronization, low-overhead messaging layers and communication protocols with OS bypass, NIC-level support, flow control mechanisms, reliability, Quality of Service (QoS), high performance implementation of emerging communication standards (such as InfiniBand, Omni-Path, 10/40/100GigE iWARP, and RoCE), supporting popular programming models (MPI, PGAS and Hybrid), high performance file systems and I/O, cloud computing environments (SR-IOV, Containers), Big Data environments (Spark, Hadoop, Mamreduce, Memcached, and HBase), and Deep Learning (CAFFE and CNTK).

A large number of state-of-the-art research projects are currently being undertaken by the group along these directions.

By participating in this seminar, the student can join in carrying out interesting research projects on the above issues and evaluate them on the large-scale experimental testbed available in the network-based computing laboratory.

Prerequisites

Permission of instructor

Materials

Papers from the literature + hands-on research projects on the experimental testbed in the Network-based Computing Laboratory.
Last Updated: August 15, 2016