Unified Middlebox Model Design and Deployment With Dynamic Resources

Abstract

Due to the expensive hardware and complex management of the traditional middlebox, a concerted effort toward the virtualized middlebox has been launched in both academia and industry. In this paper, we propose a unified middlebox model, MBBrick, which is composed of three operation modules (classifier, rewriter, and forwarder) and a control module. We then design a language, middlebox language, to describe the MBBrick process. By concatenating different modules, MBBrick can achieve the functions of different middleboxes or service chains in one single super middlebox. We design the framework for managing MBBrick and network resources. We propose the efficient algorithms to optimize the MBBrick deployment and schedule the traffic. We also consider the situation of the network failure and divide it into two types (the normal failure and the MBBrick failure). We propose the backup scheme for each MBBrick to cover the potential failure. If the failure occurs, this method can reduce the impact on the network performance. The experiment results show that MBBrick significantly reduces the processing latency and improves the network throughput.