SC385 - Optical Interconnects for Extreme-Scale Datacenters and HPC
Sunday, 03 March
17:00 - 20:00
Short Course Level: Beginner
John Shalf; Lawrence Berkeley National Laboratory, USA
Keren Bergman; Columbia University, USA
Short Course Description:
Modern Supercomputers performance is poised to soon reach the ExaFLOP mark of 1018 floating-point operations per second. Novel applications for these most powerful computers have emerged in a wide range of areas including brain modeling, the design of new materials, multi-physics simulations, and climate modeling. Large-scale data analytics and machine learning workloads from the Big Data paradigm have further pushed computing needs toward processing of large volumes of data at ever increasing speeds. To obtain such computing power, however, requires the coordinated effort of millions of processors (cores). With the extraordinary growth in parallelism, performance is increasingly determined by how data is communicated among the numerous compute resources, rather than the arithmetic operations performed. At Exascale, these challenges are daunting, as energy consumption is currently dominated by the cost of data movement, and is an increasing constraint on usable application performance. The interconnect architecture must support colossal amounts of data that is almost continuously inter-exchanged by the cores, as well as data flows present between the cores and the various memory resources. Integrated photonics offers compelling technology solutions that include high-bandwidth density interconnect and the potential for system-wide energy efficient data movement. However the insertion of photonics in future Supercomputer systems will require new architectures and systems designs. The course will include an introduction to the system organization and architectures of today’s top supercomputers as well as the emerging interconnection networking challenges. The potential applications of integrated photonics in future supercomputing and datacenters including the design, power consumption, and performance analysis will be covered.
Short Course Benefits:
This course should enable you to:
Understand how new computing technologies enable real-world applications.
Understand trends in high performance computing architecture.
Describe innovative technologies on the horizon, such as hybrid memory, optical. interconnects, multicore processors and accelerators, and petascale supercomputers.
Compare technologies and solutions for real-world applications such climate modeling, biological sciences, and materials discovery.
Point to opportunities for dramatic improvements in performance for data-movement limited applications.
Short Course Audience:
This lecture is designed to introduce students how to use parallel computers to efficiently solve challenging problems in science and engineering, where very fast computers are required either to perform complex simulations or to analyze enormous datasets. The lecture is intended to be useful for students from different backgrounds. The presenter has a strong track record of presenting similar tutorials to academic and industrial audiences, and this material will be accessible by researchers, implementers, innovators, and executives.
Keren Bergman is the Charles Batchelor Professor and Chair of Electrical Engineering at Columbia University where she also directs the Lightwave Research Laboratory (http://lightwave.ee.columbia.edu/). She leads multiple research programs on optical interconnection networks for advanced computing systems, data centers, high-performance embedded computing, and chip multiprocessor nanophotonic networks-on-chip. Dr. Bergman holds a Ph.D. from M.I.T. and is a Fellow of the IEEE and of the OSA.
John Shalf is Department Head for Computer Science at Lawrence Berkeley National Laboratory and CTO of the National Research Supercomputing Center (NERSC). John got his graduate degree in computer engineering from Virginia Tech. John first got started in HPC at the National Center for Supercomputing Applications (NCSA) in 1994. While working for the General Relativity Group at the Albert Einstein Institute in Potsdam Germany, he helped develop the first implementation of the Cactus Computational Toolkit, which is used for numerical solutions to Einstein's equations for General Relativity and which enables modeling of black holes, neutron stars, and boson stars.