Multiple cost optimization for task assignment in heterogeneous computing systems using learning automata

A framework for task assignment in heterogeneous computing systems is presented in this work. The framework is based on a learning automata model. The proposed model can be used for dynamic task assignment and scheduling and can adapt itself to changes in the hardware or network environment. The important feature of the scheme is that it can work on multiple cost criteria, optimizing each criterion individually. The cost criterion could be a general metric like minimizing the total execution time, or an application specific metric defined by the user. The application task is modeled as a task flow graph (TFG), and the network of machines as a processor graph (PG). The automata model is constructed by associating every task in the TFG with a variable structure learning automaton. The actions of each automaton correspond to the nodes in the PG. The reinforcement scheme of the automaton considered is a linear scheme. Different heuristic techniques that guide the automata model to the optimal solution are presented. These heuristics are evaluated with respect to different cost metrics