Systems are often difficult to debug and to understand. A typical way of gaining insight into system behavior is by inspecting execution logs. Manual inspection of logs, however, is an arduous process. This project helps this problem by designing Synoptic, a tool to produce summary representations of system logs. Two features distinguish Synoptic from other tools. First, Synoptic's summaries preserve key invariants mined from the log, making them more accurate. Second, Synoptic uses refinement to derive the summary, which is more efficient than traditional coarsening algorithms.

A DHT (Distributed Hash Table) is distributed key-value storage system. Although DHTs have been thoroughly researched, they are in dire need of applications. One of the reasons for this is their typically abysmal performance, loose consistency and lack of availability guarantees. This project aims to design and implement a DHT that has strict data consistency guarantees. This project is an evolution of prior work by the systems community on improving performance of DHTs, and prior work by the theory of distributed systems community on the efficiency of distributed consistency algorithms such as Paxos and its variants.

Seattle is a platform for networking and distributed systems research. It's free, community-driven, and offers a large deployment of computers spread across the world. Seattle works by operating on resources donated by users and institutions. The global distribution of the Seattle network provides users the ability to use Seattle in application contexts that include cloud computing, peer-to-peer networking, ubiquitous/mobile computing, and distributed systems. It has been used in educational settings for teaching networking and distributed systems classes.

I am also interested in analysis of social networks. This includes both quantitative analysis (e.g. member graph structures, activity patterns) as well as qualitative analysis (e.g. intensive study of activity samples, interviews). I am nourishing this interest by studying Wikipedia's policy mechanism and the interactions between wikipedia editors on discussion pages as they employ policies to arrive at consensus and make progress in their work. Our more recent work studies the span of valued work in Wikipedia by leveraging the Wikipedia Barnstars practice in which tokens of appreciation are exchanged between Wikipedia editors.

I spent six months (Summer and Fall of 2009) at MSR Asia in Beijing working as an intern in the Systems Research group under the direction of Lidong Zhou. I worked on a few projects in the mobile space, but my primary project focused on designing and developing a platform for mobile applications that can take advantage of cloud resources.

This project is based at the Max Planck Institute for Software Systems (MPI-SWS) and is part of my Summer 2007 internship work with Krishna P. Gummadi. The goal of SatelliteLab (SatLab) is to design and implement a system that improves the heterogeneity of existing internet testbeds (e.g. PlanetLab, RON) by including internet edge-nodes. Nodes located in edge-networks (e.g. residential networks) are typically unreliable, cover a wide range of software and hardware configurations, and have widely varying network connectivity characteristics. These dimensions of heterogeneity are especially important for accurately testing distributed system prototypes. Today such prototypes are developed in highly homogeneous testbed environments that hinder their readiness for realistic deployment. SatLab makes it easier to evaluate, debug, and explain the observed performance of distributed systems in the wild.

VFER is a congestion controlled transport protocol that enables client applications to define a 'functional' level of reliability with a callback reliability function. VFER has a delay-based congestion control scheme that attempts to achieve the theoretical optimum by monitoring delay variations and applying control theory to packet spacing. VFER is an experimental protocol that aims to be TCP-friendly and robust in heterogeneous network latency environments. At the moment, VFER has a C library implementation and a reliable file transfer scp-like test application. This project was initially brainstormed with Matei Ripeanu, before becoming a Google Summer of Code 2005 implementation project at Internet2 with Stanislav Shalunov as my mentor. It continued as an internship with the transport group at Internet2 and received three new students under the Google Summer of Code 2006 program, one of whom was mentored by me.

Modeling and simulation using high-performance computing are playing an increasingly important role in decision making and prediction. For time-critical emergency decision support applications, such as influenza modeling and severe weather prediction, late results may be useless. A specialized infrastructure is needed to provide computational resources quickly. SPRUCE, is a system for supporting urgent computing on both traditional supercomputers and distributed computing Grids. Currently deployed on the TeraGrid, SPRUCE provides users with ``right-of-way'' tokens that can be activated from a Web-based portal in the event of an urgent computing need. Tokens are transferrable and can be restricted to specific resource sets and priority levels. Once a session is activated, job submissions may request elevated priority. Based on local policy, computing resources can respond, for example, by preempting active jobs or raising the job's priority in the queue.

A Linux distribution effort for petascale cluster Operating Systems centered at Argonne National Labs. Currently deployed on the Compute and IO nodes of Argonne's BG\L machine, it is a massive undertaking that resulted in a GPLed codebase and promises to become a mainstream distribution for future IBM's BlueGene family clusters. My involvement with this project has been work done during the summer 2005 internship at ANL on ZOID, the ZeptoOS IO Daemon. The ZOID codebase I developed emulates the IO\Compute node syscall redirection on x86 clusters.

This effort started with a couple of classes at UChicago, and escalated into a ANL building 221 sensor-net over my summer 2005 internship at ANL. The network is two stargates with hallway mounted MicaZ motes, all running the fantastic TinyOS.