A framework for evaluating the impact of communication on performance in large-scale distributed urban simulationsTools Amponsah, Kwabena Ntim (2021) A framework for evaluating the impact of communication on performance in large-scale distributed urban simulations. PhD thesis, University of Nottingham.
AbstractA primary motivation for employing distributed simulation is to enable the execution of large-scale simulation workloads that cannot be handled by the resources of a single stand-alone computing node. To make execution possible, the workload is distributed among multiple computing nodes connected to one another via a communication network. The execution of a distributed simulation involves alternating phases of computation and communication to coordinate the co-operating nodes and ensure correctness of the resulting simulation outputs. Reliably estimating the execution performance of a distributed simulation can be difficult due to non-deterministic execution paths involved in alternating computation and communication operations. However, performance estimates are useful as a guide for the simulation time that can be expected when using a given set of computing resources. Performance estimates can support decisions to commit time and resources to running distributed simulations, especially where significant amounts of funds or computing resources are necessary. Various performance estimation approaches are employed in the distributed computing literature, including the influential Bulk Synchronous Parallel (BSP) and LogP models. Different approaches make various assumptions that render them more suitable for some applications than for others. Actual performance depends on characteristics inherent to each distributed simulation application. An important aspect of these individual characteristics is the dynamic relationship between the communication and computation phases of the distributed simulation application. This work develops a framework for estimating the performance of distributed simulation applications, focusing mainly on aspects relevant to the dynamic relationship between communication and computation during distributed simulation execution. The framework proposes a meta-simulation approach based on the Multi-Agent Simulation (MAS) paradigm. Using the approach proposed by the framework, meta-simulations can be developed to investigate the performance of specific distributed simulation applications. The proposed approach enables the ability to compare various what-if scenarios. This ability is useful for comparing the effects of various parameters and strategies such as the number of computing nodes, the communication strategy, and the workload-distribution strategy. The proposed meta-simulation approach can also aid a search for optimal parameters and strategies for specific distributed simulation applications. The framework is demonstrated by implementing a meta-simulation which is based on case studies from the Urban Simulation domain.
Actions (Archive Staff Only)
|
Tools
Tools
