PARALLEL PERFORMANCE OF STOCHASTIC ALGORITHMS
ITMO University; Junior Researcher
A. V. Boukhanovsky
ITMO University, Saint Petersburg, 197101, Russian Federation; Director
Abstract. Issues of parallel algorithm performance models construction for various task classes are discussed. Parametric approach which allows parallel speedup description in form of deterministic function is proposed. Parameters of this function are stochastic variables that characterize objective properties of algorithm and are independent of software and hardware implementation. This makes possible analytical study of dependences between performance probabilistic characteristics, algorithm and parallel architecture.
Keywords: stochastic algorithms, parallel perfomance, high-perfomance computing.