Research On Load Balancing Of Heterogeneous Wireless Networks Based On SDN

In recent years, with the rapid development of wireless communication technology, there have been a variety of wireless networks coexist. Different wireless networks have their own characteristics, and their application scenarios are quite different. The wireless networks often work independent and have no communication with each other. The radio resources between different wireless networks can’t complement each other and give full play to their strengths, which will result in a tremendous waste of resources. What’s more, with the rapid development of mobile internet, there have been a surge of the intelligent terminals, which has brought much pressure to the radio access network. Most applications on the intelligent terminals need to connect to the Internet via radio access networks, and many applications, especially multimedia applications require a large amount of wireless bandwidth. If the intelligent terminals can’t select the appropriate access point, then some access point may be overloaded and the user experience will be bad, while some other access point may have additional resources not being used.In this paper, the idea of SDN is applied to load balancing in heterogeneous radio access networks, using the load balancer to manage each radio access point. The radio access points need to report their status information to the load balancer that running load balancing algorithms, and with the information of each radio access points, the load balancer can compute if the current networks are load balanced. Whenever an application is opened, the terminal will report some requirements to the load balancer. The load balancing algorithm compute the appropriate radio access network for the terminal based on the information of network status and requirements. For the real radio access networks, a single load balancer needs to compute lots of terminal requests, the efficiency is hard to guarantee. The load balancing architecture given in this paper works in two aspects, respectively, local balance and global balance. The local controller is responsible only for the limited access points, and the global controller is connected with each local controller. With the information of each area, the global controller could make overall planning.The load balancing algorithm in the local controller is based on the utility theory, it could compute the satisfaction degree that the network parameters meet the requirements. Here the needs are divided into application needs and user needs. The application needs refer to the requirements application need to run such as network bandwidth, the user needs are the elements user care about such as network costs, power consumption, etc. The main consideration of dividing the needs is that they are quite different, e.g., the application needs are objective while the user needs are subjective, therefore their corresponding utility functions also different. When the network parameters are the same, the same kind of applications will get the equal utility value, so it is not necessary to repeat the computation. The global load controller uses the machine learning algorithm locally weighted linear regression to process the large amount of historical data provided by the local controller. By this way, we could find the potential law of the load changes in each local area, and then forecast the load, make appropriate preparations in advance.