Yuan Zhong | Bringing Clarity to Communication Systems at Scale
Assistant Professor of Industrial Engineering and Operations Research
—Photo by Ryan John Lee
Just as DNA are the building blocks of life, switches and routers are the building blocks for electronic business communication. They make it possible to connect devices like printers, computers, and servers to serve countless users, and they control how data in various forms are exchanged.
In a perfect world, any incoming request to a large, connected system would get processed instantly, and would be guaranteed a good completion time, no matter in what complicated manner the servers are wired together. But data exchange between interconnected information processing systems, such as the automation of a retail purchase that processes a payment, records an inventory change, and produces a sales receipt, rarely happens perfectly. It is often subject to unpredictable demand and resource constraints that affect efficient and timely processing. In order to bring clarity to these processing systems, researchers like Yuan Zhong, assistant professor of industrial engineering and operations research, are learning how to design simple, scalable resource allocation policies with good predictable performance.
“My research is focused on resource planning and allocation in large-scale stochastic systems such as the Internet and data centers. Better understanding and management of these systems are especially relevant today given the scale at which many of them operate,” says Zhong.
Zhong focuses on understanding performance scaling in processing networks through mathematical models. These models allow him to analyze the performance of information, communications, and manufacturing systems subjected to highly varying, unpredictable demand. He is then able to develop methods that help inform business decisions about resources needed to make those systems more efficient.
“Switched networks, for example, provide a nice mathematical framework that faithfully model many of these systems. While practically relevant, they are theoretically challenging to analyze, especially at a very large scale,” he explains.
His research has already uncovered some surprising insights. First is the use of reversibility—a classical branch of queueing network—in the design of dynamic resource allocation policies.
“One policy that we analyzed relies on a deep connection with the classical theory of reversible queueing networks, which roughly says that if we give equal share of resources to users in the system, then the system decouples into independent components, each of which is easy to analyze,” he says.
Another surprise that has come from Zhong’s research is that fairness and efficiency can be delivered together.
“We have found that fair resource allocation is efficient in that it gives optimal system scaling behavior. This is in contrast with observations from related research areas, where there is often a hard tradeoff between fairness and efficiency. For our systems, we can provide fairness and efficiency at the same time,” he says.
Zhong joined Columbia Engineering in July after completing his doctoral work at MIT, and spending one year as a postdoctoral scholar at UC Berkeley. He received the best Student Paper Award at Sigmetrics in 2012.
BA, University of Cambridge, 2006; MA, Caltech, 2008; PhD, MIT, 2012
-by Amy Biemiller