Random Approach to Optimization of Overlay Public-Resource Computing Systems

The growing need for computationally demanding systems triggers the development of various network-oriented computing systems organized in a distributed manner. In this work we concentrate on one kind of such systems, i.e. public-resource computing systems. The considered system works on the top of an overlay network and uses personal computers and other relatively simple electronic equipment instead of supercomputers. We assume that two kinds of network flows are used to distribute the data in the public-resource computing systems: unicast and peer-to-peer. We formulate an optimization model of the system. After that we propose random algorithms that optimize jointly the allocation of computational tasks and the distribution of the output data. To evaluate the algorithms we run numerical experiments and present results showing the comparison of the random approach against optimal solutions provided by the CPLEX solver.