An Adaptive Reservation Scheme for Effective QoS Provisioning in the Cloud SDN Stack

The virtualized cloud hosts multiple applications each with unique traffic characteristics necessitating QoS provisioning for its finite network resource. While hypervisor and network based virtualization techniques provide traffic isolation in multitenant cloud environments, there is a gap in their interworking & integration for effective end-to-end SLA maintenance. Recent solutions in cloud network provisioning through reservation practices have accomplished some degree of success. However they are oblivious of the entire application communication path & don't provide the requisite SLA. In this work, we develop a reservation methodology that is aware of the complex hierarchy of the cloud network. Furthermore, given the volatility of traffic generated and received by cloud tenant applications, we employ a general class of dynamic score models of time-series to develop a predictive algorithm integrated with the bandwidth provisioning framework. This solution is resident in the application layer of the SDN stack as a controller. Our methods have been tested in the OpenStack environment that integrates a layered cloud network in a real-life test-bed. Our predictive methods provide a 15% increase in accuracy over current methods while able to provide effective SLA maintenance across clients with unique QoS requirements.