Research Projects @ Grad School (2008-2013)

Xorbas (XORing Elephants: Novel Erasure Codes for Big Data)

  • Regenerating codes specially designed for data centers are being implemented over Hadoop HDFS
  • Based on Facebook's implementation of HDFS-RAID
  • About 2x reduction in network utilization and disk I/O during file repairs.
  • Advised by Prof. Alex Dimakis.
  • Source code available at https://github.com/madiator/HadoopUSC
  • Appeared in VLDB 2013.
  • Visit the project page.

Network Coding for Vehicular Networks

  • Network Coding applied to Vehicular Networks to minimize the delay in content retrieval.
  • Obtained theoretical upper bounds on delay and showed regions where Network Coding performs better than naive distribution strategy.
  • Simulated on real taxi traces to show the improvement.
  • Continued the work at General Motors to test on real vehicles.
  • Advised by Prof. Bhaskar Krishnamachari, Prof. Alex Dimakis and Dr. Fan Bai.
  • Appeared in Infocom-Mini 2012 and journal submission under progress.

VANETSim: A Vehicular Network SimulatorJan 2012 -- Present

  • A Java based simulator primarily designed to study coded and uncoded content distribution strategies in vehicular networks.
  • The goal is to simulate a city-wide network of vehicles exchanging content with one another.
  • Advised by Prof. Bhaskar Krishnamachari in collaboration with General Motors R&D

Twitter Retweet Dynamics Apr 2010 -- Aug 2011

  • Using data collected from Twitter, we showed an interesting trend between the number of retweets received and the probability of retweeting by Twitter users.
  • Work was presented at the 2nd Annual Annenberg Symposium and was invited to present again at the 3rd Annual Annenberg Reception.
  • Future work involves characterization of Tweets based on their space-time properties, spam detection in Twitter etc.
  • Advised by Prof. Antonio Ortega.

Implementation of MapReduce Feb 2010 -- Apr 2010

  • Implemented MapReduce (a framework for distributed processing) on USC's High-Performance Computing and Communications (HPCC) cluster consisting of thousands of nodes, as part of a course.
  • Used it to study huge network datasets.
  • Course (Computer Communications, CS551) advised by Prof. John Heidemann.

Other Projects

  • Implementation of Nachos, a software that simulates a small OS, involving process management, memory management, interprocess communication, fault tolerance etc. for a course on Operating Systems.
  • Flash Scheduling - proposed and analyzed a new scheduling algorithm in a multi-user communication system with varying number of users (term project for Computer Communications course).

We don't beat the Reaper by living longer. We beat the Reaper by living well - Randy Pausch