Creating a multi-agent simulation environment for evaluating the economic models of cloud providers

Abstract

At present much research is being carried out into simulating the performance of Cloud computing services and infrastructures. One area which appears to have been overlooked so far is the human aspect. Human usage patterns clearly have a large impact on the levels of service demand at speci c time points. Agents have long been employed to simulate complex real-world en- vironments in urban environments, computer games and other application areas. This thesis examines whether similar techniques can be developed for modelling the usage patterns of cloud users and their behaviours. The pur- pose of this work is to explore and simulate the requests of cloud users for IAAS services. In order to achieve this the rst phase was to gather realistic data. After gathering as much realistic data as was available, a multi-agent simulator was implemented which had three layers of hierarchy. These in- cluded planning, task utilization and allocation e ciency and a multi-agent simulation environment was implemented using these features. To validate the approach simulated cloud users behaviour was compared with real-world usage patterns which were gathered and then analysed using quantitative analysis. The results con rm that the approach is approximately 80% ac- curate and therefore it would appear that the simulator is realistic enough to be useful for generating usage patterns which can then feed into existing models in order to provide a more realistic picture of an IAAS environment.