Prof. Dr. Frank Bellosa, Dr. Jan Stoess

Abstract:

Cloud computing environments play an important role in the today’s computer era. Therefore computing clouds need to be able to deploy several hundreds to even thousands of virtual machines in a scalable way and to provide adequate migration and backup strategies. To achieve this, cloud providers need to use efficient mechanisms to deploy and migrate VMs and need to avoid potential bottlenecks like a central storage system. In this thesis we present our approach to enable fast deployment and migration in an HPC environment that exploits the tight interconnects of a Blue Gene/P supercomputer. Our approach uses main-memory-based distributed storage and runs directly inside the hypervisor that allows our approach to directly access the VM’s memory. By using the remote DMA capabilities of BG/P’s interconnects we are able to minimize load at the storage nodes and reduce the potential of a bottleneck they pose. Although the main purpose of our approach is to provide snapshotting, it is able to copy any data into the servers.

BibTex:

@studythesis{gutekunst12vmdeployment,
author = {Manuel Gutekunst},
title = {Enabling fast VM deployment and migration in tightly interconnected HPC cloud environments},
type = {Study Thesis},
address = {System Architecture Group, Karlsruhe Institute of Technology (KIT), Germany},
month = jan # "31",
year = 2012,
url = {http://os.ibds.kit.edu/}
}