Analysis of job scheduling algorithms for heterogeneous multiprocessor computing systems

The problem of scheduling independent jobs on heterogeneous multiprocessor models (i.e., those with non-identical or uniform processors) with independent memories has been studied. Actually, a number of demand scheduling nonpreemptive algorithms have been evaluated, with respect to their mean flow and completion time performance criterion. In particular, the deterministic analysis has been used to predict the worst-case performance whereas simulation techniques have been applied to estimate the expected performance of the algorithms. As a result from the deterministic analysis, informative worstcase bounds have been proven, from which the behaviour of the extreme performance of the considered algorithms can be well predicted. However, relaxing some or a combination of the system parameters then, our model corresponds to versions which have already been studied. (i.e. the classical homogeneous and heterogeneous models or the homogeneous one with independent memories). For such cases, the proven bounds in this thesis either agree or are better and more informative than the ones found for these simpler models.. Finally, the analysis of the worst-case and expected performance results reveals that there is a high degree of correlation in the behaviour of the algorithms as predicted or estimated by these two performance measurements, respectively.