“In recent years, many techniques have been developed to improve the performance and efficiency of data center networks,” writes Professor Theophilus “Theo” Benson of Brown CS, as he explains the significance of his most recent paper on Deep Learning. This paper was recently recognized by ACM SIGCOMM, and was a runner-up at a large workshop just this year. SIGCOMM is the ACM’s professional forum for the discussion of topics in the field of communications and computer networks, with a particular focus in systems engineering and architectural questions of communication.
Currently, Theo’s work is precisely in this field as his specialty is in solving practical networking and systems problems, with a focus on software defined networking, data centers, clouds, and configuration management. This paper centers on how many data center networking techniques (routing, topology augmentation, energy savings) share design and architectural similarities. He presents a framework for developing general representations of network topologies using deep learning to solve a large class of data center problems, and to simplify the process of configuring and training deep learning agents. This framework was used to implement a DeepConf-agent that tackles the data center topology augmentation problem, and resulted in performance comparable to the optimal solution.
This paper follows a natural evolution of Theo’s research, as he works on understanding and designing techniques for data analysis. His paper “Network Traffic Characteristics of Data Centers in the Wild” won the best paper award at IMC 2010. This was followed by his paper “MicroTE: Fine Grained Traffic Engineering for Data Centers,” which involved work with Microsoft on domain specific algorithm to improve the performance of data centers, and his paper “YTrace: End-to-End Performance Diagnosis in Large Cloud and Content Providers,” which explored domain specific techniques for diagnosis at Yahoo data centers. His research has progressed from large scale analysis, to domain specific heuristics, to application of statistical techniques, before finally evolving to machine learning.
For more information, click the link that follows to contact Brown CS Communications Outreach Specialist Jesse Polhemus.