End-to-end performance of multi-core systems in cloud environments

Multi-core systems are widespread in all types of computing systems, from embedded to high-end servers, and are achievable in almost all public cloud providers. The sophistication of the hardware and software architectures make the performance studies of such systems very complicated. Further complexity is introduced by the virtual environments which are the basis of all clouds paradigms. While there have been several studies concerning the performance of multi-core systems considered stand alone, few of them are focused on the end-to-end performance of these systems when accessed through virtualized platforms. In this paper we describe the results obtained with experiments on both Amazon EC2 and VirtualBox platforms. The experiments are performed with some of the DaCapo benchmarks and with IOzone. The objective is to explore at a high abstraction level how the interference between the characteristics of the applications and those of the architectures impact on the performance that users of multi-core systems experience. We also designed some expressions that, although the high-level of abstraction and the low complexity, have a good precision with regard to the performance prediction of the overall system. We think this is a first step toward understanding the end-to-end performance that a multi-core system is able to provide when accessed through a cloud platform.