Taming Orchestration of Complex Services in the Cloud

The growing complexity of Cloud technologies is revealing several limitations of traditional resource- management processes that cannot deal properly with the scale of workloads and dynamic resource requirements of constantly evolving applications and services. As a consequence, delivering acceptable QoS levels while improving resource utilization requires significant expertise and manual configuration that compromise operational costs and the agile on-boarding of new virtualized services. Cloud complexity requires more sophisticated management techniques designed to 1) reduce the con- figuration burden through automation, and 2) leverage abstractions to account for workload and system heterogeneity while satisfying end-to-end service performance. We propose an automation framework that executes a collection of data-driven algorithms to create a holistic representation of the service that models workloads mixes and resources requirements including component dependencies and the impact of heterogeneous stacks. We have evaluated our framework using a telecommunication service that handles multiple classes of workloads with different hardware requirements and support a variety of devices. Our results show that with our approach underprovisioning can be eliminated, in the best case or, be reduced at least to a 33 - 50%. This is translated to a KPI improvement of at least a 20%.