Scheduling Divisible Loads in Parallel Processors

Divisible loads are ones that can be arbitrarily and independently partitioned across a number of processors for the purpose of parallel processing. Applications include the processing of very large linear data files where the data may be numerical, graphical or video. Optimal scheduling for divisible loads boosts speedup and decreases solution time on the order of up to 50%. Optimal scheduling is most effective when the computation and communication intensity of the load is similar.

In this talk we look at the basic principles and simplicity of optimal divisible load scheduling. This topic has been the a sub-area of research for a dozen years. We'll also look at some lessons learned from this research.

Top of Page