Design and analysis of control schemes for soft real-time computation systems

Abstract

This thesis studies the management of task execution percision in dynamic soft real-time computer systems which utilise imprecise computations. A Notice Generation (NG) approach is utilised to decide on task execution precision. We analyse four different NGs, with these four NGs at our disposal, six methods are proposed to deal with transient overload in the system. The overload is modelled by a significant increase in the mean arrival rate of tasks to the computation system. Results show that this method provides significant performance benefits over single scheme methods that ignore transient overloads.