Srikant seeks to exploit variability in social, cloud networks
Kim Gudeman, Coordinated Science Lab
- ECE Professor R. Srikant is part of a team designing systems to optimize heavy-tailed distribution in networks.
- He is developing a model for making inferences about the ability of a social network's power user to influence connections.
- Srikant is also examining resource allocation for cloud computing.
In a computer network, file sizes vary from Internet pages that are a few kilobytes to movie files that are many gigabytes. Likewise in a social network, some users have several dozen friends or followers, while others have several thousand.
While the traditional view is that wide variability in networks poses a problem—video files hog bandwidth and delay even small operations—ECE Professor R. Srikant believes that it is possible to exploit those variations for positive outcomes. He is among researchers at six universities that will explore how to design systems to optimize heavy-tailed distribution in networks. The project, “Multivariate Heavy -Tail Phenomena: Modeling and Diagnostics,” will be funded by an Army Research Office Multidisciplinary University Research Initiative (MURI) for five years at $6.25 million.
“We hope to create a model that would enable you to understand or make inferences about behavior based on an observation,” said Srikant, the Fredric G. and Elizabeth H. Nearing Endowed Professor of Electrical and Computer Engineering and a researcher in the Coordinated Science Lab.
In a computer network, the idea of exploiting variability is already reasonably well understood. However, it’s less clear how the concept would work in social networks and cloud computing systems – the focus of this project.
In a social network such as Facebook, some people have a large number of friends, while others have just a few. Marketers who are looking to maximize their advertising dollars may want to place an ad on the Facebook page of a user with a large network, since they will likely influence more people than others with fewer friends. Srikant is working to create a model that would enable organizations to make inferences about a power user’s ability to influence his or her connections without actually knowing who the user’s friends are.
In cloud computing, variability is key in allocating resources. As in the case of the Internet, cloud providers want to ensure that small jobs are not delayed by getting behind very large jobs in the queue. Srikant’s group will develop a methodology for understanding how to schedule and route jobs, taking into consideration the wide variety of the number and size of jobs at any given time.
In addition to this grant, Srikant will be studying related problems in cloud computing through another NSF grant, “Resource Allocation in Clouds: A Stochastic Modeling and Control Perspective.” The grant, which will be a total of $450,000 for three years, will be split with Professor Lei Ying, Srikant’s former PhD student, at Arizona State University.
The research could have applications outside of social networks and cloud computing. The military could potentially use it to understand or infer relationships based on behaviors. It may also help companies build better business models. A company like Family Video, for example, could make better predictions about what type of movies to have in stock, likely the most popular blockbusters, and what type of movies to offer online, such as indie movies that have a smaller but devoted audience base.
“This work could really be applied to any number of networks,” Srikant said. “It will allow a more targeted approach to business, defense and other areas.”