Abstract des Autors

Cloud-based services revolutionize how applications are designed and provisioned in more and more application domains. Operating a cloud application, however, requires careful choices of configuration settings so that the quality of service is acceptable at all times, while cloud costs remain reasonable. We propose an analytical queuing model for cloud resource provisioning that provides an approximation on end-to-end application latency and on cloud resource usage, and we evaluate its performance. We pick an emerging use case of cloud deployment for validation: sports analytics. We have created a low-cost, cloud-based soccer player tracking system. We present the optimization of the cloud-deployed data processing of this system: we set the parameters with the aim of sacrificing as least as possible on accuracy, i.e., quality of service, while keeping latency and cloud costs low. We demonstrate that the analytical model we propose to estimate the end-to-end latency of a microservice-type cloud native application falls within a close range of what the measurements of the real implementation show. The model is therefore suitable for the planning of the cloud deployment costs for microservice-type applications as well.