Characterizing fault tolerance in genetic programming

摘要: Evolutionary Algorithms (EAs), and particularly Genetic Programming (GP), are techniques frequently employed to solve difficult real-life problems, which can require up days or months of computation. One approach reduce the time solution is use parallel computing on distributed platforms. Distributed platforms prone failures, when these large and/or low-cost, failures expected events rather than catastrophic exceptions. Therefore, fault tolerance recovery often become necessary. It turns out that Parallel GP (PGP) applications have an inherent ability tolerate failures. This quantified via simulation experiments performed using failure traces from real-world platforms, namely, desktop grids (DGs), for two well-known problems. A simple technique then proposed by PGP better different, high, rates seen in different