In this paper, we model cloud systems and the user interactions with the cloud provider using the UML2Cloud profile.In general, users request virtual machines according to their needs, but they can also subscribe to the cloud provider and wait to be notified when the requested resources are not available.In this case, users indicate a maximum subscription time, so once this time elapses without being notified, users leave the system unattended.In this paper, then, we present an exhaustive research study to measure how the user subscription times affect the overall system responsiveness.In this study, three different cloud configurations are analyzed.Each cloud processes several workloads, which are generated using two distribution functions for the user arrivals, namely a normal and a cyclic normal distribution.The purpose of this study is to find out the inflection point for the waiting time of the users, from which the cloud responsiveness and its performance do not improve.The obtained information is therefore useful for the cloud provider to improve the configuration of the cloud.

Cloud providers face the challenge of managing large amounts of heterogeneous resources in real time. It is usually very costly to conduct experiments with real cloud systems. Therefore, tools to analyze and evaluate cloud scenarios and experimental studies are very useful for them. In this paper, we model cloud systems and the user interactions with the cloud provider using the UML2Cloud profile. In general, users request virtual machines according to their needs, but they can also subscribe to the cloud provider and wait to be notified when the requested resources are not available. In this case, users indicate a maximum subscription time, so once this time elapses without being notified, users leave the system unattended. Thus, we present an exhaustive experimental study to measure how the user subscription times affect the overall system responsiveness. To this end, four different cloud configurations are analyzed, and the workloads for these studies are produced by using three distribution functions for the user arrivals, namely, a uniform, a normal, and a cyclic normal distribution. Furthermore, we also analyze the cloud performance with a workload obtained from a real trace. The purpose of this study is to find out the inflection point for the waiting time of the users, from which the cloud responsiveness and its performance do not improve. The obtained information is, therefore, useful for the cloud provider to improve the configuration of the cloud.