Study of SDN Controller Locations in Network Environments
Date Issued
September 22, 2021
Type
Μεταπτυχιακή Διπλωματική Εργασία
Abstract
The emergence of the public and private cloud services resulted in on demand access to infrastructure, applications and other resources. For instance the the big data applications requires parallel processing of massive data on thousands of servers. These servers must be able to communicate with each other, to produce the final result. This communication requires a huge amount of network bandwidth. This amount of data raised many challenges to the existing TCP/IP network architecture. Software Defined Networking (SDN) has been proposed as a new network architecture to tackle these challenges.
The SDNs are capable of managing the network resources in a centralized manner. They allow dynamic and efficient network configuration to increase the network's performance. This increase in performance is accomplished by disassociating the forwarding process from the routing process. The network's function is decoupled into two distinct planes: the control plane and the data plane. The control plane is in charge of managing the network's resources efficiently, while the data plane is responsible for forwarding the packets from source to destination.
A common use case of SDNs is to facilitate backbone networks. Backbone networks consist of large number of devices in the network and relatively long distances between nodes. The long distances of between nodes raises a constraint that the controller of a SDN could not be placed in a bad location, i.e. the controller has minimum distances between a handful of switches and long distances between the remaining devices. Instead, the controller must be placed at a location that its distance from the devices of the network is minimum. This problem is the 1-median problem.
There are numerous ways to tackle this problem the most common is to formulate the problem in an integer linear problem and then solve it using heuristics. This study presents an approach that incorporates elements of algebraic graph theory. The proposed approach uses the spectral characteristics of the distance matrix, in which rows and columns are indexed from the nodes of the graph and is populated with the shortest distance between each pair of nodes. More specifically, the proposed method uses the eigenvector of the distance matrix that corresponds to its largest eigenvalue, this eigenvector is also named principal eigenvector of the distance matrix. The selection of median is done by finding the element of the principal eigenvector that is closer to zero.
The evaluation of the proposed method is performed over various cases and number of nodes in random tree graphs. Its aim is to select the optimal solution of the median problem applied to SDNs. A program in Python programming language is developed in order to facilitate this evaluation process. This program is responsible to create the random tree graphs, produce the distance matrix and then to compare the acquired solution to the actual median.
The proposed method is firstly evaluated on random tree graphs with each edge weight equal to one. The proposed method is not able to find the optimal solution in some cases of these graphs. The proposed method is also evaluated on random tree graphs that its edges have random weights. The results show that the proposed method can find the median of a graph in most cases even in large-scale topologies.
Subjects