Extreme Data-Intensive Scientific Computing

摘要

Scientific computing increasingly involves massive data; in astronomy, observations and numerical simulations are on the verge of generating petabytes. This new, data-centric computing requires a new look at computing architectures and strategies. Using Amdahl's law to characterize architectures and workloads, it's possible to use existing commodity parts to build systems that approach an ideal Amdahl machine. Modern science is approaching the point where novel computational algorithms and tools, combined with computational thinking, will become as indispensable as mathematics. This trend is being driven by the fact that scientific datasets across a range of fields are doubling in size every year, thus creating complex challenges for traditional analysis techniques. Analyses of the information contained within these datasets have already led to revolutionary breakthroughs in fields ranging from genomics to high-energy physics, encompassing every scale of the physical world. Much more remains undiscovered.