Cloud Application HA Using SDN to Ensure QoS

Abstract

Users expect cloud applications to be highly available with minimum service disruption. Some of the cloud applications also have Quality of Service (QoS) requirements. The High Availability (HA) module needs to consider QoS requirements while placing or failing over the application, application's components and its replicas. In this paper we propose a new QoS module in the SDN controller. This module creates QoS queues with certain minimum bandwidth on Open Flow based switches on the route between components requiring certain minimum network bandwidth. Our approach is similar to 'Aggregation of RSVP for IPv4 and IPv6 Reservations' (RFC 3175). DiffServ based approaches reserve bandwidth for an entire class of flows and do not cater for requirements from individual applications. Our approach reserves bandwidth on a per-flow basis while overcoming the well-known scalability problem of IntServ. Also, the QoS queues are dynamically configured based on the current requirements of the applications. The QoS module is also responsible for ensuring QoS when topology and route changes. The HA module will interact with SDN controller (which includes our QoS module) to ensure QoS requirements of cloud applications. We also tried an intermediate approach where the QoS flows are directed to a DiffServ queue with a fixed rate (mentioned in RFC 3175) and the reservation of the queue is reconfigured only when the total reservation for the QoS flows exceed the initial reservation. We demonstrate the feasibility and performance aspect of our approach from experiments using Mininet.