Efficient and transparent virtualization mechanism for volunteer computing

• Main Author: Anjomshoa, Mohammadfazel
• Format: Thesis
• Language: English
• Published: 2014
• Subjects: QA75 Electronic computers. Computer science
• Online Access: http://eprints.utm.my/id/eprint/78267/1/MohammadfazelAnjomshoaMFSKSM20141.pdf; http://eprints.utm.my/id/eprint/78267/; http://dms.library.utm.my:8080/vital/access/manager/Repository/vital:79546

Virtualization is a technology that introduced long time ago but it has emerged in the last decade as a viable and novel solution for assembling a complete operating system (guest OS) on top of hosting machine (host OS). It has changed computing in many aspects with its unique features like easy deployment of guest OSes, migration, checkpointing and sanboxing. Public resource computing project like SETI@home, Roesetta@home and others that are powered by volunteer resources can benefit from virtualization characteristics in both project developing and volunteers points of view. However wide-scale deployment of virtualized environment for desktop grids impacts on the host performance as virtualization functionalities imposes Central Processing Unit (CPU) and memory overhead into the host environment. Virtualization adoption imposes additional download bandwidth on volunteers machine. This thesis aims to propose an efficient approach to adapt virtualization into volunteer computing platform. The proposed virtualization mechanism is implemented on BOINC. It uses VirtualBox to establish virtualized environment. In order to reduce resource overhead, a centralized virtual machine undertakes the execution process which is created by a symlink virtual machine image file. Evaluation results demonstrate that the proposed virtualization approach, improves BOINC performance in terms of CPU and memory overhead both Random Access Memory (RAM) and storage notions. The proposed mechanism reduced the CPU overhead by 96.17%, 206.5%, 316.85, and 429.47% when executed single job, two jobs, three jobs and four jobs in parallel respectively. In the case of memory overhead, the proposed virtualization mechanism improved the storage overhead by 95.5%, 194.60%, 220.75%, and 286.43%, and declined the RAM overhead by 0.00% , 100%, 200%, and 300% when scaled up from executing single job to four jobs respectively. The proposed virtualization mechanism reduced considerably resources overhead which were occupied by virtual machine environment and depicts the possibility of adapting virtualization functionality into the volunteer computing environments with the acceptable additional overhead.