Platform virtualization is an essential component of Infrastructure as a Service (IaaS) type cloud computing offerings. It enables customers to gain access to computing resources in a logically isolated and transparent way. However, due to the resource sharing nature of platform virtualization and the additional limited observability of public cloud platforms, there are major open questions in the field regarding Quality of Service (QoS) assurance. Specifically, there is no widely accepted methodology yet that how transient or trend-like resource-allowance failures can be detected at runtime, modeled and their QoS effects characterized.
In my BSc thesis I describe an experimental system based on open-source components that I have used to benchmark service level metrics on private and public cloud platforms.
I performed two major measurement campaigns. In the first, I performed controlled resource allowance fault injections on a private cloud platform. In the second, I ran the benchmark setup on Amazon EC2, a widely used public cloud platform. My work also included experiments with the novel „mystery shopper” concept.
My results show that the effect of resource allowance interferences on service levels is significant enough to get dedicated support for acquiring the necessary empirical input for the application deployment design phase.