| PhD Viva

Name of the Speaker: Ms. Navya Vuppalapati (EE17D056)
Guide: Prof. Venkatesh TG
Venue: ESB-244 (Seminar Hall)
Online meeting link: https://meet.google.com/aob-qwdm-gio
Date/Time: 24th May 2024 (Friday), 10:00 AM
Title: Resilient Software Defined Networks:Traffic Characterization and Candidate Selection under Link Failures

Abstract :

In traditional communication networks (TCN), routers and switches are connected using wired or wireless mediums to transmit packets. Each router or switch in TCN operates independently, making its own decisions regarding packet forwarding. This decentralized nature of TCN poses challenges when modifying the network structure or introducing new services, as changes must be made to each network element individually. Additionally, the rapid growth of 5G and Internet of Things (IoT) networks has increased the demand for higher quality of service (QoS) and dynamic network management, putting additional strain on the existing TCN infrastructure. To address these challenges, Software Defined Networking (SDN) has emerged as a new paradigm. SDN provides a centralized network view, enabling the implementation of new services without the need to reprogram each forwarding element or switch individually. In an SDN environment, switches are designed to be non-intelligent, meaning they simply forward packets based on instructions received from a centralized controller. This centralized control and programmability make SDN more flexible and adaptable to changing network requirements, allowing for easier network management and the rapid deployment of new services.

The dynamic nature of SDN has made it a preferred choice over TCN. However, this dynamic nature introduces non-stationarity in network traffic, which can have significant implications for network performance and management. In this thesis, our primary objective is to characterize SDN traffic and analyze its behavior under dynamic conditions. To achieve this, we employ the Augmented Dickey Fuller (ADF) test to study the non-stationary aspects of SDN traffic. Additionally, we develop a queuing model that can effectively capture the non-stationary conditions of SDN networks. We use the Pointwise Stationary Fluid Flow Approximation (PSFFA) technique to solve for the average queue length at both the controller and the switch in such dynamic scenarios. To ensure the reliability and accuracy of our findings, we validate our analytical results through extensive simulations. Through our work, we highlight the significance of considering the non-stationary behavior of traffic in the design and analysis of SDN systems.

Next, we address the candidate selection problem in hybrid Internet Protocol (IP)/SDN, specifically focusing on Single Link Failures (SLFs). Our objective is to minimize the number of SDN candidates necessary to safeguard against SLFs in hybrid IP/SDN networks. To accomplish this, we introduce an innovative algorithm called Minimum Candidates Selection (MCS). The MCS algorithm considers all feasible shortest paths between pairs of nodes in the network. By leveraging this comprehensive approach, we can effectively optimize the candidate selection process. To evaluate the performance of the MCS scheme, we assess two key metrics: average repair path length (ARPL) and average maximum link utilization (AMLU). By implementing the MCS algorithm on various real-world and random network topologies, we demonstrate its superiority over existing algorithms.

Later, we evaluate the performance of a hybrid IP/SDN network in handling single bidirectional link failures. We investigate two CSAs to identify the candidate switches: the Greedy CSA and the Minimum CSA. Initially, we focus on these CSAs with the consideration of a single shortest path between any two nodes within the network. Subsequently, we expand our study by incorporating all the shortest paths between every pair of nodes. Furthermore, we address the challenge of deploying additional candidate switches when there is flexibility in the budget constraints (i.e., the number of candidate switches). To assess the performance of the aforementioned CSAs, we evaluate their ARPL and AMLU across various real-world network topologies.

Finally, we investigate the critical scenario of multi-link failures (MLF) in a hybrid IP/SDN network. To effectively address this challenge, we propose four CSAs based on a non-shortest path approach. Additionally, we introduce a CSA designed specifically for network scaling or expansion, which optimizes the allocation of candidate switches to safeguard newly established links. Furthermore, we present a budget constraint-based CSA that assists in selecting the most suitable candidate switches within a given budget constraint. To evaluate the performance of the hybrid IP/SDN network with MLFs using the proposed CSAs, we assess the ARPL across various real-world network topologies.