Resource optimization of TCAM-based SDN measurements via diminishing-points autodetection

Abstract

Network measurement is an important tool for network managers and operators since it provides the information needed to carry out different management tasks. However, because of the rapid increase in data link speeds and the volume of traffic carried by modern networks, the availability of system resources dedicated to network measurement has always been the main limiting factor in developing modern measurement solutions. The emergence of Software Defined Networks in recent years has inspired the development of new measurement solutions that takes advantage of the capabilities offered by this new paradigm such as programmability and central management [9]. Current SDN enabled measurement solutions are able to orchestrate the execution and resources allocation of network-wide measurement tasks but still falls short in their ability to recognize efficient operating points (amount of allocated resources) for running tasks and often results in inefficient resources utilization. In this paper we provide a novel resource allocation method that continuously estimates the resources-accuracy relationships for running tasks, infers their individual points of diminishing-returns and uses the resulting value of each task as it's target point of operation in order to achieve a more efficient resource utilization. In contrast to existing work, our proposed method maximizes the return on used system resources and results in lower drop rates of running tasks as shown by our simulations.