Abstract: Over the next decade the largest computer systems will enter the era of Peta operations per second and climb to Exa operations per second. Although the basis of this growth will be an increasing number of processor cores per chip and an increasingly non-uniform memory access architecture, the number of chips and operating systems nodes will need to grow rapidly to achieve performance goals. The inevitable consequence of increasing component counts is more parts that can fail, higher failure rates, more concurrent failures and more effort devoted to coping with and recovering from failures. In this talk I will review historical data on failure rates in supercomputers to project future failure rates, review growing limitations on traditional fault tolerance strategies for supercomputers based on high-speed checkpointing to parallel storage systems, and look in depth at the increasing failure issues in storage components to chart new challenges in the largest storage systems and new reliability mechanisms to address these challenges.

Bio: Garth Gibson is a Professor of Computer Science and Electrical and Computer Engineering at Carnegie Mellon University, is the co-founder and Chief Technology Officer at Panasas Inc., a leading provider of parallel storage systems, is the co-director of the Institute for Reliable High Performance Information Technology sponsored by Los Alamos National Laboratory, and is the Principal Investigator of the US Department of Energy's Petascale Data Storage Institute in the Scientific Discovery Through Advanced Computing program (www.pdsi-scidac.org). Gibson's PhD research at the University of California, Berkeley, developed the Redundant Arrays of Inexpensive Disks (RAID) taxonomy, and earned him a 1992 dissertation award from the Association of Computing Machinery (ACM), a 1998 Test of Time Award for most influential paper ten years earlier from the ACM Special Interest Group for Management of Data (SIGMOD), a 1999 Reynold B. Johnson Information Storage Award from the Institute of Electrical and Electronics Engineers (IEEE), and a 1999 Allan Newell Award for Research Excellence from Carnegie Mellon University. Gibson founded the Parallel Data Laboratory, academia's premiere storage systems research center, at Carnegie Mellon University and in this lab led the Network-Attached Storage Device research that became the basis of the newly standardized T10 Object-based Storage Devices (OSD) command set for SCSI storage. At Panasas (www.panasas.com) Gibson led the development of the ActiveStor Parallel Storage Cluster, used in supercomputers and Fortune 500 HPC systems to provide scalable performance with visionary reliability improvements to RAID storage. Gibson also co-instigated the Internet Engineering Task Force's nearly complete open standard for parallel Network File Systems (NFSv4.1), and directs Panasas' contributions to the development of open source NFSv4.1 in the Linux operating system. Gibson was a co-founding member of Technical Council of the Storage Networking Industry Association (SNIA), is a co-founding member of the Steering Committee of the USENIX Conference of File and Storage Technologies, and has given ten keynote speeches to academic and industrial conferences around the world, including the 2007 International Conference on Parallel Processing in XiAn, China.