Optimization of Shared-Memory Multicore Systems Using Game Theory and Genetic Algorithms on Cellular Automata Lattices

A main problem in multi-core architectures is the runtime management and the allocation of shared resources, such as the shared memory. This paper presents a model of the memory resources allocation, in specific the on-chip shared memory, in order to achieve higher performance based on the basic concepts of game theory and the iterated spatial prisoner's dilemma game, with the help of adaptive computational tools like cellular automata and genetic algorithms. The paper evaluates the proposed scheme using multi-core applications that are based on the MapReduce programming framework. The performance evaluation and the simulation results show that the allocation of shared resources based on game theory and genetic algorithms, can improve, up to 10%, the overall performance of the multicore architectures.