As we move closer to extreme scale computing, applications face numerous challenges that must be addressed before they can deliver breakthrough science. I/O is one of the major challenges many of the applications are facing at the extreme scale. My team has led the efforts into creating a new paradigm for data management on HPC platforms, which utilizes a Service Oriented Architecture to combine data-in-transit techniques with extreme performance I/O, and in situ data processing. This talk will present the evolution of this vision, from its inception to assembling a collaborative team, to breakthrough science delivered using our software platform. In addition to 45 publications over the last three years, our software stack has gained widespread adoption in the HPC community, and is being used in over 24% of the allocated time on the Oak Ridge Leadership Computing Facility. In the final part of my talk, I will layout the future of this vision, extending the overarching end to end goal of our HPC efforts with a new twist of combining it with cloud storage infrastructure, enabling remote collaboration and driving new insight into extreme scale data.
Scott Klasky is a Senior research scientist, and the end to end task lead, in the NCCS and CSM divisions of Oak Ridge National Laboratory. Scott received his Ph.D. from the University of Texas at Austin, and has been working in the area of high performance computing for the past 23 years, working with applications in relativity, astrophysics, combustion, and fusion. Scott is the leader of the eSiMon and ADIOS projects, and is an expert in high performance, and data intensive computing. Scott is currently funded by the Oak Ridge Leadership Class Facility, SciDAC, DOE Advanced Scientific Research combustion co-design center, NSF, OFES, and NASA. ADIOS is open source and available from http://www.nccs.gov/user-support/center-projects/adios/.