About me
I am an Assistant Professor in the Department of
Computing and Mathematical Sciences at Caltech where
I am a member of the Rigorous System Research Group (RSRG, pronounced "resurge").
I also participate actively in two centers: the Center for the Mathematics of Information (CMI), which focuses on the intersection of Applied Mathematics and Computer Science, and the Social and Information Sciences Laboratory (SISL, pronounced "sizzle"), which focuses on the intersection of Economics and Computer Science.
My research
My research focuses on using mathematical models to provide insight into the design of computer systems. I have two main halves to my work: one which focuses on scheduling and resource allocation in computer systems and one which focuses on network economics.
Broadly, my particular research focus can be described as: "Better design through modeling and measurement." More specifically, some examples of topics that I'm focusing on these days are:
-
Making ICT sustainable
These days, everyone has heard the statistics about how much of an energy hog ICT has become: The emissions of a server are nearly that of a car! The electricity usage of data centers is growing 12 times faster than that of the US as a whole! While the last decade has led to significant improvements in energy-efficiency across IT, there is still a long ways to go to be truly sustainable. Our work in this area... -
The economics of the smart grid
Over the coming decade, the electricity network will undergo the same architectural transformation that the telephone network has recently gone through to become more sustainable, more interactive, and more autonomous. However, there are huge engineering and economic challenges in making this transformation possible. In fact, unlike in the case of communication networks, the market structure and engineering are inherently intertwined in the electricity grid, which necessitates a new architectural theory for guiding this transformation. Our work in this area... -
Network economics
These days it is almost impossible to study networking without considering economic incentives. From net neutrality, to the design of P2P systems, to hot potato routing, understanding the economic incentives in networks has become crucial. Our work in this area... -
The role of computation in economic models
One of the fundamental tasks for research at the intersection of CS and Economics is understand how to incorporate "computation" into classic economic theories. Toward this end, over the last decade, the task of understanding the computational complexity of economics models has been a cornerstone of the field and an understanding has emerged that many standard economic models are "hard" in the worst case. However, this "algorithmic" perspective on computational hardness differs from the "empirical" perspective with which many economists view the models, i.e., that a model is simply a way to explain reality (data). Our work seeks to understand: "do computational restrictions have empirical consequences for economic models?" . Our work in this area... -
Exploiting the structure of social networks
Our understanding of the structure of social networks (and other complex networks) has grown dramatically over the last decade and from this understanding has emerged nearly `universal' properties such as small world properties, heavy-tailed degree distributions, etc. Our research looks at how these properties can be exploited to solve problems that, without such structure, would be intractable. Our work in this area... -
Heavy tails versus light tails: Is it possible to design for both?
``Bad'' events in heavy-tailed workloads are most often the result of ``catastrophes'', e.g., one enormous job. In contrast, ``bad'' events in light-tailed workloads are most often the result of ``conspiracies'', e.g., the combination of many small jobs. These "catastrophe" and "conspiracy" principles guide the design of schedulers and controllers in each setting. However, ideally one would like to be able to design schedulers/controllers to be robust, i.e., optimal in both heavy and light tailed settings. So, the question becomes: "Is it possible for schedulers/controllers to avoid both catastrophes and conspiracies? Our work in this area... -
Non-cooperative, cooperative control
Distributed cooperative control problems are everywhere, and are especially important in wireless and sensor network applications; however designing robust distributed control protocols is especially challenging. We are studying an inherently robust approach to distributed control that is based on formulating the distributed agents as selfish players in a non-cooperative, engineered game. Our work in this area... -
Queueing games
Queueing theory has a long and storied history, but throughout this history, nearly all work has assumed that the workload is a fixed and non-interactive. In reality, as performance changes users will react and so the workload will also change. Modeling such interactions requires a mixture of queueing theory (to model the system) and game theory (to model the users). Once the strategic users are considered, new perspectives on classic results emerge. Our work in this area... -
Fairness and scheduling
Modern designs often improve user response times by giving priority to small job sizes. But, this leads to worries about whether large job sizes receive fair performance. So, we need to ask: How much starvation/unfairness is caused to large jobs by prioritizing small jobs? Our work in this area...
For more information about these and other current projects see my publications.