A study in grid simulation and scheduling

Abstract

Grid computing is emerging as an essential tool for large scale analysis and problem solving in scientific and business domains. Whilst the idea of stealing unused processor cycles is as old as the Internet, we are still far from reaching a position where many distributed resources can be seamlessly utilised on demand. One major issue preventing this vision is deciding how to effectively manage the remote resources and how to schedule the tasks amongst these resources. This thesis describes an investigation into Grid computing, specifically the problem of Grid scheduling. This complex problem has many unique features making it particularly difficult to solve and as a result many current Grid systems employ simplistic, inefficient solutions. This work describes the development of a simulation tool, G-Sim, which can be used to test the effectiveness of potential Grid scheduling algorithms under realistic operating conditions. This tool is used to analyse the effectiveness of a simple, novel scheduling technique in numerous scenarios. The results are positive and show that it could be applied to current procedures to enhance performance and decrease the negative effect of resource failure. Finally a conversion between the Grid scheduling problem and the classic computing problem SAT is provided. Such a conversion allows for the possibility of applying sophisticated SAT solving procedures to Grid scheduling providing potentially effective solutions.