Speakers
Dr
Antun Balaz
(Institute of Physics Belgrade)Mr
Dusan Vudragovic
(Institute of Physics Belgrade)Mr
Vladimir Slavnic
(Institute of Physics Belgrade)
Description
Monte Carlo (MC) applications are inherently parallel and computationally-bound. In a distributed computing infrastructure, once MC application is initiated, it can be usually executed independently, i.e. with almost no interprocess communication. Such behavior allows for a usage of any kind of computing resources, but requires an integrated framework to simplify application management, and even most importantly, proper usage of the underlining parallel random number generator and reproducibility of numerical results. Here we present newly developed toolkit for management of large-scale Monte Carlo simulations on various kinds of computing resources, which includes management of parallel random number streams utilized by individual processes, as well as data gathering and post-processing functionality.
Primary authors
Dr
Antun Balaz
(Institute of Physics Belgrade)
Mr
Dusan Vudragovic
(Institute of Physics Belgrade)
Mr
Vladimir Slavnic
(Institute of Physics Belgrade)
