Abstract

ORNL has now procured a supercomputer with a peak performance of over 1 petaflop, with over 300 TB of memory, 150K processors, and 10 PB of storage. One of the most daunting tasks of running simulations on this computer is with the associated scientific data management tasks, such as parallel I/O and workflow automation. In this talk I will present two middleware systems that we have worked with, and present the new research challenges that remain in each task. In particular, I will focus on the Adaptable I/O System (ADIOS), and the Kepler workflow system. We present many exciting research areas that our end to end team are working on, including: data staging/offloading, code coupling, in situ visualization, fast reading of log based file formats, autonomic workflows, and fast writing of metadata rich output from 100K processors.

Speaker

Dr. Klasky is currently a senior research scientist at Oak Ridge National Laboratory, and head of the end to end task in the scientific computing group at the National Center Computational Sciences. Dr. Klasky has 20 years of experience in research and development of middleware for use in high performance computing, and is a author/co-author of over 40 papers in the field of physics and high performance computing. Dr. Klasky received his Ph.D. in physics from the University of Texas at Austin, and then went on to Syracuse University, and then the Princeton Plasma Physics Laboratory. Dr. Klasky is currently on 4 DOE SciDAC projects for work on data management techniques for petascale simulations.