Article Preview
TopIntroduction
SDN (Sahu & Hungyo, 2018; Singh et al., 2019) is an emerging network that is capable of transforming the traditional network architecture due to its support for programmable networks and open protocols as discussed in (Nunes et al., 2014). The SDN paradigm bifurcates the traditional network architecture with distributed forwarding data plane and a centralized control plane. The network's “brain” is the SDN Controller which is used to communicate with the hardware architecture and controls traffic on the network. The custom applications are developed in the SDN controller to support a variety of responsibilities like load balancing; anomaly detection, and dynamic centralized routing decision.
The data center is the backbone of IT infrastructure and consumer service-based applications. The exponential growth of mobile computing and the application-based environment caused the data centers to carry a huge amount of traffic with unpredictable behavior. As per the work presented by Index, (2016); Kumar et al., (2021), “The amount of annual global data center traffic in 2015 is already estimated to be 4.7 ZB and by 2020 will triple to reach 15.3 ZB per year”. Therefore, Data Center requires better traffic management for efficient and optimized resource utilization.
Traffic Engineering (TE) consists of methods devised to optimize the performance of the data network. TE uses static or dynamic analysis of traffic for better management to avoid congestion and better utilization of bandwidth (Kumar & Tiwari, 2021). Elephant flows are presented by Kandula et al. (2009), which came in less frequency but consume huge bandwidth. This causes sudden network congestion. The detection of such flows can increase the efficiency of the overall system since the post-detection the flows be rerouted to load balance the network.
The Datacenter (DC) traffic is of two types, user traffic, and server traffic. User traffic consists of data by the services provided to the user containing external traffic. Whereas, the server traffic is generated due to the inter-server communication. The DC network requires significant aggregate bandwidth. Typically, it has a tree-like architecture made up of routing and switching components. In this kind of implementation, only a small portion of the total bandwidth is made available to the edge network.. Therefore, a topology such as the Fat-tree topology provides better distribution of bandwidth and ensures network reliability.
A comprehensive view and control of the entire network are provided by SDN. It has been successfully implemented in different DC networks like Google B4 (Jain et al., 2013). SDN provides network programmability and dynamic policy updates. SDN-based traffic engineering solutions are effective in DC networks as discussed by Ian et al. (2014). In the present paper, a method is provided for elephant flow that is directly implemented as a module in the floodlight controller, which is further elaborated in the paper. The proposed method requires polling from the network switches and identifies the elephant flow and the cheetah flows in the network. The proposed method uses a dynamically computed threshold value for flow detection and its value is computed based on the network traffic characteristics.
The rest of the paper is organized as follows. The details of the recent works suggested by various researchers are discussed in the “Background” section. Subsequent section, after background the section, provides the “Proposed Methodology”. The result analysis has been discussed in “Results” section and finally, the paper is concluded in the “Conclusion and Future Works” section.