Evolutionary task scheduling in static and dynamic environments

This paper presents an analysis of the behavior of an evolutionary algorithm in the context of scheduling tasks in static and dynamic distributed computing environments. The dynamic character of the computing environment is simulated by randomly marking some resources as unavailable. Some memory based and diversity preserving mechanisms were investigated in order to asses their ability to deal with the dynamic nature of the environment. The experimental results suggest that using information from the schedule evolved for the previous state of the environment is beneficial if the difference between the sets of available machines at two consecutive stages is not larger than 12%. When more changes occur in the list of available machines from one scheduling event to the next one it is usually better to start from scratch the construction of a new schedule.