Optimal dynamic redistribution of divisible load in distributed real-time systems

The problem of optimizing the balancing of processing load originating at the various sites of an interconnected set of heterogeneous processors of a real-time system is examined. The distributed load is assumed to be homogeneous and arbitrarily divisible with no particular affinity to individual processors. The heterogeneous processors may or may not have front-end subprocessors, and are characterized by their load processing speeds and I/O speeds pertaining to the transmission of excess load or reception of additional load. The objective of load redistribution, which is carried out during runtime, is to minimize the job execution completion time consisting of the sum of processing lime and I/O time. Depending on the value of total load, the optimal load redistribution is found to be either unique or to admit a multiplicity of solutions. In the first case, all processors complete processing in the same amount of time, which is equal to the minimum job completion time. In the second case, the infinite set of optimal redistributions is represented by load allocations for the individual processors that can range over specified intervals, and the completion limes for the different processors are generally not equal