A novel four‐tier software‐defined network architecture for scalable secure routing and load balancing

• Published in: International journal of communication systems Vol. 35; no. 1
• Main Authors: Hongvanthong, Sisamouth, Chunlin, Li
• Format: Journal Article
• Language: English
• Published: Chichester Wiley Subscription Services, Inc 10.01.2022
• Subjects: Algorithms; Analytic hierarchy process; Ant colony optimization; Clustering; Communications traffic; Computer architecture; Controllers; Cryptography; Curves; elliptic curve cryptography; fuzzy analytical hierarchy process; Load balancing; Machine learning; multicontroller; Optimization algorithms; Particle swarm optimization; Recurrent neural networks; Response time; scalable secure routing; Shortest-path problems; Software-defined networking; Switches
• DOI: 10.1002/dac.5020
• ISSN: 1074-5351

Summary Software‐defined networking is an emerging paradigm for supporting flexible network management. In the traditional architecture for a software‐defined network (SDN), the controller commonly uses a general routing algorithm such as Open Shortest Path First (OSPF), which chooses the shortest path for communication. This may cause the largest amount of network traffic, especially in large‐scale environments. In this paper, we present the design for a novel SDN‐based four‐tier architecture for scalable secure routing and load balancing. In Tier 1, user authentication is conducted using elliptic curve cryptography (ECC); this avoids unnecessary loads from unauthorized users. In Tier 2, packet classification is performed based on the packet characteristics using the fuzzy analytical hierarchy process (fuzzy AHP), and packets are placed into three individual queues. In Tier 3, scalable secure routing is achieved by selecting the optimal path using the improved particle swarm optimization and ant colony optimization algorithms. With these optimization algorithms, we can adaptively change the number of users, the number of switches, and other parameters. In Tier 4, the recommended secure cluster (multicontroller) management is accomplished using an algorithm that employs modified k‐means clustering and a recurrent neural network. Deep reinforcement learning (DRL) is also proposed for updating the controller information. Experimental results are analyzed using the OMNeT++ network simulator, and the evaluated performance displayed improvement over a variety of existing methods in terms of response time (50% to 60%), load (55%), execution time (3.2%), throughput (9.8%), packet loss rate (1.02%), end‐to‐end delay (50%), and bandwidth consumption (45%). We proposed novel four‐tier architecture in which load balancing and security considered in each tier. First, all nodes (users) are authenticated using ECC, which balance the network by avoiding extra traffic from unauthorized users. Second, packets are classified into three classes by fuzzy AHP. Third, routing is established among switches in secure way by improved ACO and PSO algorithms. Finally, clustering is performed to avoid single failures by modified k‐means and RNN. DRL is proposed for controllers’ information updated.