Modeling large populations of full-sized virtual machines using minimal virtual instances

Abstract

This thesis proposes the use of minimal virtual machines to model larger populations of instances in different environments, and investigates if a cloud environment is able to take the populations even further. Finally the thesis wants to investigate if minimal virtual machines are suitable to host custom application stacks and are able to compete with full-sized virtual machines. As virtualization technology has achieved increased popularity the recent years virtual machines are now used by many businesses, institutions and consumers for different purposes. Full-sized virtual machines are large, and demand considerable amounts of computing resources from the Cloud Resource Pool. This project was able to significantly reduce the size of virtual machines and the amount of computing resources required to host them. The smallest virtual machine accomplished in this project had a size of merely 1.5MB allowing a population of almost 500 times, or at least two orders of magnitude, larger than one standard-sized Ubuntu Server instance. Custom written software was also created for each type of virtual machine for the purpose of simulating real-world CPU usage patterns. Several population sizes of minimal virtual machines were deployed and tested in Hypervisor-on-Hardware and Hypervisor-in-Cloud labs to compare their behavior and performance in different environments.