Distributed computing as a service with ARC middleware

Barbara Krašovec (2016) Distributed computing as a service with ARC middleware. MSc thesis.

Preview

PDF Download (3522Kb)

Abstract

Scientific computing has evolved considerably in the past years. Scientific applications became more complex and require an increasing number of computing resources to perform well on a large scale. Grid computing became widely used and is the chosen infrastructure for many scientific calculations and projects, although it demands a high learning curve. Computing and storage resources in grid are limited, heterogeneous and often overloaded. Heterogeneity is not present only in the hardware setups, but also in software composition, where configuration permissions are limited. This heterogeneity hardens the portability of scientific applications. Usage of cloud resources could eliminate those constraints. In cloud, resources are provisioned on demand and can scale up and down, scientists can easily customize their execution environments in the form of virtualization. The interest for grid and cloud integration is big, since the new infrastructure would enable a relatively simple migration of scientific applications from grid to cloud (or vice versa). The utilization of resources would be more efficient and it would provide scalability, sustainability and reliability of the hybrid infrastructure. In the IaaS domain, Advanced resource connector (ARC) middleware could have a role of distributed high throughput batch computing as a service. ARC has an ability to facilitate distributed computing across organizations. It is strongly committed to open standards and interoperability and in a good position to enable users and research communities to access new resources and new platforms using existing and well-established interfaces and standards. It could provide a way to seamlessly evolve its interfaces to enable existing users and communities to take full advantage of new resources and technologies as they become available. Some integration models exist, but usually only correspond to a specific use case. In this thesis we introduce a new model of grid and cloud integration which enables users to benefit from additional private and public cloud resources via ARC, with no modifications of the code, workload or execution scripts on their side, by setting up a virtual grid cluster on demand in automated way. The solution can be globally used, on all private and public clouds, it is scalable and corresponds to the cloud bursting paradigm.

Actions (login required)