Abstract

The main focus of this research is in the area of adaptive scheduling for heterogeneous distributed systems. Given an unreliable, non-dedicated set of processing and communication resources, a scheduler is required to allocate tasks to processors. No information about the state of the system, which can vary over time, or the tasks to be processed, is known in advance and thus must be estimated dynamically. Current schedulers do not adequately address this dynamism. To address this, a property estimation method is presented, which utilizes a k-Nearest Neighbours algorithm, a smoothed average and an analytical benchmark. These estimated properties are then used by two different scheduling techniques, which make less restrictive assumptions than the current state-of-the-art methods. A multi-heuristic evolutionary method utilizes a genetic algorithm and eight simple heuristics to efficiently allocate tasks to processors. A deterministic method utilizes the error inherent in estimating the properties of the system and the execution time of tasks, to allocate tasks to processors. The algorithms have been implemented on a real-world heterogeneous distributed system with up to 150 processors. A set of real-world problems from the areas of cryptography, bioinformatics, and biomedical engineering were used as a test set to measure the effectiveness of the scheduling algorithms. Experiments have shown that both methods achieve better efficiency than other state-of-the-art heuristic algorithms. Finally, a low memory distributed reconstruction application for large digital holograms is presented, which has significantly increased the size of holograms that can be reconstructed, over the previous state-of-the-art.