Shreyasee Mukherjee

This work analyzes Linear Program based algorithms for centralized tunnel routing in a software defined network. The proposed algorithm splits MPLS tunnels across K-shortest paths between each pair of end-points in order to optimally route traffic and improve the system capacity. Results were obtained based on a 50 node realistic ISP backbone network. Although 10-shortest paths are available for each pair of nodes, the figure on the left shows 97% of the traffic does not require to be split.

Link: Summer internship final presentation slides

Collaborators: Gagan Chaudhuri, John Klincewicz (AT&T Labs, NJ)

In this project, we considered the problem of designing and implementing distributed versions of K-SVD, a widely used machine learning algorithm for distributed clusters and supercomputers. Implementation was done and tested on Spark and analytical calculations on cost vs. throughput tradeoff of distributing data across machines was performed.

Links: Coming soon!

Collaborators: Yuanxi Li, Maryam Dehnavi (Rutgers)