Reliability management framework for softwarized networks

Abstract

The Software-Defined Networking (SDN) technologies promise to enhance the performance, reliability, and cost of managing both wired and wireless network infrastructures, functions, controls, and services (i.e., Internet of Things). However, centralized reliability management in Softwareization architecture poses both scalability and latency challenges. Significantly, the current OpenFlow Discovery Protocol (OFDP) in SDN induces substantial scalability, accuracy, and latency hurdles due to its gossipy, centralized, periodic, and tardy protocol. This dissertation proposes a novel reliability management framework, which efficiently orchestrates different reliability monitoring mechanisms over SDN networks and synchronizes the control messages among various applications. The proposed framework facilitates multiple discovery frequency timers for each target over different stratum instead of using a uniform discovery timer for the entire network. It supports many common reliability monitoring factors for registered applications by analyzing offline and online network architecture information such as network topologies, traffic flows, virtualization architectures, and protocols. The framework consists of a high availability registration platform (HARP) and the topology-aware reliability management (TARman) and Bug Detection, Debugging, and Isolation (BuDDI) protocol facilities. The reliability management framework is implemented on both Ryu and Cisco’s OpenDayLight (ODL) controllers. Extensive Mininet experimental results validate that framework significantly improves discovery message efficiency and makes the control traffic less bursty than OFDP with a uniform timer. It also reduces the network status discovery delay without increasing the control overhead. Our reliability management framework also proposes a novel network reliability cost model to ensure that the SLA covers customer service impact and damage. We classify network outages and calculate their effect on the network services to formulate a cost-based model. Besides, we have performed evaluations using various campus network outage scenarios. The proposed cost-based model enables customers to identify the service impact of unplanned network outages to their networks instead of entirely depending on the service provider’s data.