High Performance Grid Computing for Optimization Problem (Mathematics and Algorithms of Optimization)
克樹 藤沢 · Kyoto University Research Information Repository (Kyoto University) · 2002
Grid computing has recently received much attention as apowerful and in- expensive methodology for solving large numerical problems that an existing single CPU cannot process.Ninf is agrid computing infrastructure which enables us to easfly access computational resources including hardware and software library distributed across awide area network.The basic Ninf system employs aclient-server model, where the server and client machines are connected via alocal area network or the Internet.We have been applying the Ninf system to optimization problems and polynomial systems of equations.Among others, Takeda et al. reported that the Successive Convex Relaxation Method (SCRM) on aLAN setting Ninf system can deal with some large size Quadratic Optimization Problems through numerical experiments.To solve larger size QOPs, how- ever, we need more computing resources, and we implemented ahighly parallel SCRM which utilizes several PC clusters connected via the Internet.In this talk, we discuss grid computing for optimization problems through some numerical experiments of the SCRM and polynomial systems of equations.