HCW Keynote

Summary form only given. Modern computing platforms provide huge amounts of computational power - the top supercomputers contain more than 100,000 cores, and volunteer computing grids gather millions of processors. Resource selection becomes critical for enabling scientific computing applications to fully harness such platforms: the more resources at our disposal, the more difficult the art of selecting which ones to enroll in the execution, and of mapping the right task onto the right machine. Resource optimization is quite difficult because modern platforms have new, and hard to manage, characteristics: they contain heterogeneous processors (multicores, GPGPUs, ...); they may be distributed on a very large scale, which can significantly impact communications; they may be volatile and even unreliable. In addition, several conflicting objectives (execution time, reliability, energy consumption, ...) must be considered simultaneously. In this talk, I will provide a few examples of resource selection strategies, and (hopefully!) demonstrate their importance. These examples will cover a broad spectrum, from pleasantly parallel applications (independent tasks) to tightly-coupled scientific kernels.