Challenges of Future High-End Computing

×ØÖÖ BLOCKINØØ With the recent achievement of one teraflop/s (½¼ ½¾ flop/s) on the ASCI Red system at Sandia National Laboratory in the U.S., many have asked what lies ahead for high-end scientific computing. The next major milestone is one petaflop/s (½¼ ½½ flop/s). Systems capable of this level of performance may be available by 2010, assuming key technologies continue to advance at currently projected rates. This paper gives an overview of some of the challenges that need to be addressed to achieve this goal. One key issue is the question of whether or not the algorithms and applications anticipated for these systems possess the enormous levels of concurrency that will be required, and whether or not they possess the requisite data locality. In any event, new algorithms and implementation techniques may be required to effectively utilize these future systems. New approaches may also be required to program applications, analyze performance and manage systems of this scale. º½ ÁAEÌÊÇÇÍÍÌÁÇAE In December 1996, a sustained rate of one teraflop/s (also written as ½¼ ½¾ floating-point operations per second) was achieved on the Linpack benchmark, using the ASCI Red system at Sandia National Laboratory in New Mexico, U.S.A. While no one has yet demonstrated a sustained performance rate exceeding one teraflop/s on a real scientific or engineering application, it is expected that this also will be achieved soon. The next major milestone is a performance rate of one petaflop/s (also written as ½¼ ½½ floating-point operations per second). It should be emphasized that we could just as well use the term " peta-ops " , since it appears that large scientific systems will be required to perform intensive integer and logical computation in addition to floating-point operations, and completely non-floating-point applications are likely to be important as well. In addition to prodigiously high computational performance, ¾¿