Title: Increasing effectiveness in virtualized infrastructure
Author: Tobias Lindinger
e-mail: lindinge@nm.ifi.lmu.de
Partner: LMU
Supervisor: Heinz-Gerd Hegering
Committee:
Year of start: 2007
Year of end: 2009
Funding institution: Ludwig-Maximilian-Universität München

At a first glance virtualization techniques offers great benefits to computing centers. Indeed such infrastructures can be operated without great efforts using latest virtualization techniques, but migration towards a virtual infrastructure is both cost and time intensive. Normally the old physical infrastructure is not able to serve a virtual infrastructure, because only few but highly productive servers are needed whereas many but low productive servers are available. So a first step towards virtualization is to organize new hardware components. One challenge regarding this is that no common guidelines are available telling how to calculate the number and dimensions of the needed systems. Indeed some manufacture specific tools, monitoring the physical infrastructure over several month do exist, but the monitored results have to be analyzed by the manufacture itself - no criteria for analysis is available for public. Comparing several products is so nearly unpossible, because each evaluated product must be monitored for several month. Normally, due to financial reasons one is interested in the smallest infrastructure possible, but if the infrastructure is scaled to little it is overcharged and can not perform it’s tasks.

The intention of this PhD thesis is to identify parameters influencing performance of virtual infrastructures in a theoretical and empirical manner. In a first step static parameters as for example type of the virtualization approach, number and type of virtualized systems regarding resource requirement and efficiency are analyzed. By means of this analysis an information basis can be created enabling or at least simplifying resource planing at migration time. In a second step dynamic factors of influence to performance, such as configuration items of virtual machines are examined. Results of this step can be used to calculate a more efficient resource distribution within the virtual machines and more efficient distribution of virtual machines on physical servers. This is the basis to develop real-world solutions of resource schedulers that can be used for migrating hardware to virtual machines and virtual machines to hardware depending from the situation and costs incurring by the scheduling.