My research lies in the broad area of applied probability and stochastic systems, with applications in data centers, cloud computing, and privacy-preserving data analytics. Such applications are the backbone of many ever-developing technologies, especially the emerging big-data technology. Enormous challenges are presented by these new technologies, including scalability to large sizes, coordination of data and computation, ultra-low latency, economic efficiency, etc. The goal of my research is to address these challenges, provide a clear understanding of fundamental limits of systems, and build theoretical foundations for designing new architectures and algorithms.
Stochastic systems with applications in data centers and cloud computing
I am interested in characterizing fundamental limits of large-scale computing systems that address emerging demands from big-data analytics, and designing algorithms with optimality on throughput and latency performance. My research in this area focuses on the following aspects.
Latency characterization. Latency is a crucial performance metric in technologies nowadays, especially in interactive applications such as real-time machine learning, online transactions, and video streaming. Amazon has calculated that a page load slowdown of just one second can cost $1.6 billion in sales each year. My research in this area provides tight characterizations of latency in large-scale computing systems, enabling identification of performance bottlenecks and informing optimal designs of scheduling and resource allocation algorithms.
Coordination of data and computation. Big-data analytics has led to a paradigm shift from computation-centric computing systems to systems where data plays an equally, if not more, important role. Data and computation are closely correlated in present computing systems. For example, data-parallel frameworks such as MapReduce/Hadoop and Spark associate a computing task with a chunk of data, where the computation can be executed only when both the associated data and a computation slot on servers are available. My research takes the lead in addressing the coordination of data and computation, and designs scheduling algorithms with rigorous theoretical guarantees on performance optimality.
Large-scale regimes. To accommodate the growing computational demand, modern data centers are scaled up in size, consisting of tens of thousands of servers. Such large-scale systems have attracted great attention both from practice and from theoretical research. My research investigates the scaling behavior of performance with respect to the size of system infrastructures, including the number of servers, the number of communication links, etc. My research further explores new large-scale regimes to capture the large volume of data. In particular, I study jobs whose sizes expand as the system scales, which models the trend that computational jobs are processing larger and larger volumes of data.
I work on data privacy and its intersection with other areas including game theory, information theory and statistics. Our recent focus is a new, market model that we envisage for collecting private data where data subjects (individuals) retain full control of their privacy. I aim to understand the economic fundamentals of collecting private data and design optimal incentive mechanisms.