Enabling engineering applications for the cloud on the mOSAIC platform

Jernej Južna (2013) Enabling engineering applications for the cloud on the mOSAIC platform. MSc thesis.

Preview

PDF Download (3394Kb)

Abstract

This Master’s Thesis investigates the problem of porting existing computationally and data intensive engineering applications to the Cloud. Cloud computing technologies can offer a lot of advanced possibilities, such as ubiquity, scalability and fault tolerance to engineering applications. However, existing applications do not take advantage of Cloud computing technologies. Successful porting necessitates the knowledge of a wide spectrum of technologies as well as development time for adaptation of existing applications. For this reason many engineering applications that would be of interest to the general public, are used by a limited number of people within research and development departments. In the context of this work, we developed an effective and general purpose programming code that can be used by engineers to port their applications to the Cloud with minimal changes of their existing applications. Our experiments focused on applications developed in MathWorks’ Matlab and the open source mOSAIC Platform as a Service (PaaS). The mOSAIC PaaS can be installed on top of an arbitrary Infrastructure as a Service (IaaS) provider and thus eliminates the vendor lock-in problem. The developed procedure was implemented and extended with a compiler, which facilitates data exchange between the end users and the resulting Web service in the Matlab programming language. The performance of the overall solution was tested against two applications: a benchmarking application for the estimation of the Matlab performance and a civil engineering application for analysis of structures under static loading, for which we used two different Cloud providers - Amazon EC2 and Eucalyptus. The results indicate that the developed programming code facilitates successful porting of both applications to the Cloud. At the same time numerical stability and adequate performance are achieved. The developed programming code is useful also in the case of parallel usage of the Cloud enabled applications by an increasing number of users as both applications exhibit linear scale-up. Finally, we considered the possibility for parallel implementation of the application for analysis of structures under static loading. The aim was to exploit the possibilities offered by distributed computing environments, such as the Cloud. The application was rearranged and the testing results show that approximately double speed-up can be achieved by a four times increase in the consumption of Cloud resources. The low speed-up may be attributed to the nature of the employed algorithm and to the low throughput of the input/output system, which is caused by the mOSAIC platform computing stack. Some other engineering applications would certainly gain more with parallelization.

Actions (login required)