The Approach to Flow Management in Virtual Computational Environment for Up-to-Day Telecom Networks

Abstract

AbstractData streams that are transmitted over communication networks are growing very quickly, this is facilitated by the progress of the service such as IoT, M2M, etc. The provision of high-quality communication services is increasingly influenced by the efficiency of performing calculations in the data centers of telecom operators, whose resources are limited. SDN technologies make it possible to redistribute transmitted data streams in networks to reduce their load, but they require more computing resources. Telecom operators’ networks not only transmit significant amounts of information, but also process very large information flows in their computing nodes, including virtual ones. Controlling the transfer process with a large number of threads becomes a bottleneck. To service flows in SDN, an application (at the application level) call is required, which entails overloading the computational resources involved in processing incoming flows. The main approach for dealing with the computing resources overload in such systems is load balancers that should distribute the load between computing nodes efficiently. However, such approaches lead to increased latency as the flow peak rate increases and there is a computing resources limitation. In this chapter, to reduce the time for making decisions when load balancing, «the endless train» method is proposed. This method, instead of analyzing the input flow state and simultaneously the resources state, analyses the state of the computational resources only to make a decision regarding needed resources based on the current task requirements. This allows reducing the time for making a decision on the choice of a server serving the input flow. To test the efficiency of the proposed method, an implementation scheme was developed using MS Azure. The process of dynamically deploying additional virtual servers (virtual machines) to handle threads in the event of overload was tested. The testing results show the effectiveness for overload prevention but the computational recourses usage is increased.KeywordsSoftware defined networkComputation in SDNLoads balancer