R. Srikant wins two INFOCOM awards
Running parallel to the scenic Charles River in Oregon is a series of massive, football-field sized warehouses. These nondescript buildings don’t seem like much, but inside are hundreds of racks of servers, humming rhythmically as they process data from millions of users. This is Google’s cloud computing system.
This model was highly effective at first, but is starting to lose steam as exponentially higher amounts of data come in to the servers, and the task of sampling servers to send the data starts taking too long.
In a new paper, CSL Professor R. Srikant (ECE) along with Arizona State University Professor Lei Ying (MS '03, PhD '07) and Ying’s graduate student, Xiaohan Kang, has come up with another method to speed this mechanism up.
In Srikant’s new system, tasks are automatically gathered in batches of 500, then each task samples a different random server. Out of the 500 randomly sampled servers, the tasks start sorting themselves into the servers with the least load, then gradually up to higher loads until all of the tasks are sorted. It’s like 500 shoppers getting together at the checkout line, and assigning people to the smallest lines until everyone has a spot at the checkout queue.
This is the genius of Srikant’s algorithm, which he calls “water-filling,” to evoke the idea of filling up an entire container or batch before moving on the next one. By collecting data in batches and dispatching tasks to servers on a larger scale, as opposed to individually assigning each one a job, Srikant has achieved double the effectiveness of the power-of-two model with his new algorithm.
INFOCOM, IEEE’s conference focusing on research into computer and data communication networks, has given Srikant’s work the Best Paper Award, and also named him for its Career Achievement Award. INFOCOM presented him with both of these at its annual conference in Hong Kong.
“R. Srikant is one of the most prominent, prolific and influential researchers in the INFOCOM community,” reads his award on the INFOCOM website. “Throughout the years, R. Srikant has made significant contributions in many areas of networking, including TCP congestion control, wireless scheduling, P2P modeling, social and economic networks. He is best known for developing a theoretical foundation that unifies optimization, control and queueing theories for the design and analysis of distributed resource allocation algorithms for communication networks.”
Srikant primarily researches and models computer networks and cloud computing systems. He models protocols that govern the transmission of data, especially the methods that machines use to route and sort data when it arrives at the receiving end, and is concerned with coming up with ways to efficiently and fairly deliver data in an Internet with a finite amount of resources and a burgeoning amount of data-hungry users.
“Mostly my work tends to be mathematical in nature,” Srikant said. “I come up with models, try to understand the way networks function, then come up with ways to improve current protocols and methods that could be used in creating new ones.”
Apart from developing his water-filling algorithm, Srikant has made numerous contributions to the field of data networks, for example, modelling Transmission Control Protocol, or TCP: one of the Internet’s fundamental governing structures for sending and receiving data, based on a set of rules that makes the receiver acknowledge it has received a packet before the host sends the next section of information.
Srikant also created the first widely cited model to describe peer-to-peer networks like BitTorrent. Though Srikant has had enough success to be recognized for two INFOCOM awards simultaneously, he said he believes working at Illinois has contributed to these successes.
“I’m very happy to be recognized, but I think the award itself is a function of being in a place as prestigious as Illinois, and having truly stellar faculty and students to work with,” Srikant said.