End-to-End Performance Prediction for Selecting Cloud Services Solutions

In cloud computing, in order to select or recommend the best service solutions to end users, the end-to-end QoS requirements (e.g. response time and throughput) have to be computed. A typical cloud solution is a combination of multiple component services such as IaaS, SaaS, PaaS, etc. In a simplified case, there could be two components- software services and infrastructure services. The software service alone can satisfy end user´s functional requirements (e.g. business objectives); however, the end-to-end QoS requirements require a collaboration of the multiple components at multiple cloud layers. In this paper, we consider the multilayered cloud architecture for computing the end-to-end performance values for cloud solutions. We propose a new method for measuring cloud component services similarity and predicting the end-to-end performance values of cloud solutions. In this method, the historical performance data of cloud component services is used based on users´ past invocations. To evaluate our method and show its effectiveness, series of experiments are conducted. The experimental results demonstrate that our cloud multi-layers based method produces better prediction accuracy than other prediction approaches that consider one cloud layer.