Performance analysis of different virtualization architectures using OpenStack

Abstract: Cloud computing is a modern model for having on demand access to a pool ofconfigurable resources like CPU, storage etc. Despite its relative youth however, it has already changed the face of present-day IT. The ability to request computing power presents a whole new list of opportunities and challenges. Virtual machines, containers and bare-metal machines are the three possible computing resources which a cloud user can ask from a cloud provider. In the context of this master thesis, we will discuss and benchmark these three different deployment methods for a private OpenStack cloud. We will compare and contrast these systems in terms of CPU, networking behavior, disk I/O and RAM performance in order to determine the performance deterioration of each subsystem. We will also try to empirically determine if private clouds based on containers and physical machines are viable alternatives to the traditional VM based scenario.To achieve these goals, a number of software suites have been selected to act as benchmarks with the aim of stressing their respective subsystem. The output of these benchmarks is collected and the results are compared against each other. Finally, the different types of overhead which take place between these three types are discussed.