In the context of computational grids, a metascheduler is the service responsible for scheduling jobs across many geographically distributed processor clusters. Typically, these software systems are complex and difficult to understand, resulting in initial designs which are ad hoc and suboptimal. This paper shows how a formal design methodology can be to used to better understand the relationships between metascheduler parameters, and thereby to achieve a well-designed metascheduler. Using a Plackett-Burman design, the methodology is demonstrated in the design of a knapsack grid metascheduler. The design is performed using efficiency and delay as the target variables.
Daniel C. Vanderster, Nikitas J. Dimopoulos, Randa