Distributed Simulation Cloud Architecture for Experimentation (DICE)

Abstract

Distributed Simulation (DS) is a method in operational system analysis that has gained interest due to its claimed benefits, including model reusability and interoperability. DS allows the exploitation of geographically distributed resources such as equipment and people. However, the cost of high-performance computing resources, technical skills, and special training required to design, develop, and use DS is an ongoing concern. These are the long-standing challenges that have prevented the broader adoption of parallel and distributed simulation technology. Cloud computing offers an alternative approach to address these issues using the pay-as-you-go economic model, eliminating considerable investments in the required hardware and software. DS has the potential to benefit Modelling and Simulation (M&S). Nevertheless, relatively limited attention has focused on developing a framework and deployment architecture to enable analysts to run DS experimentation on the cloud. A more in-depth study is needed to understand how modellers will run cloud-based DS and how the cloud platforms will perform with variant parameter inputs. The literature established that DS development is a complex process and requires expertise with immense courage to undertake. This thesis investigated how the cloud can be used to connect geographically distributed federates to analyse operational systems. To achieve that, a deployment architecture is proposed and experimented with potentials benefit modellers. Furthermore, a development methodology is proposed to guide analysts at every step of the cloud-based distributed simulation (CBDS) implementation - from concept to cloud execution. The experimental results indicate that it is feasible to connect and run geographically distributed simulation using cloud infrastructure. The research further finds that running a federation on a single cloud performs differently than federation execution on multiple cloud platforms. The significant differences are primarily attributed to how each cloud service provider handles network traffic and the overall communication overheads found on the Internet. This research has contributed to the CBDS approach and focussed more on analysing operational research systems by less technical modellers. The principal contributions of this work include a proposed scalable CBDS deployment architecture - DIstributed simulation Cloud Architecture for Experimentation (DICE). DICE becomes the foundation of this research, providing technical specifications and guiding analysts on deploying DS on various cloud platforms.