PhD Candidate - Computer Science
Graduate Research Assistant - DEPEND@CSL
University of Illinois at Urbana-Champaign
I am a graduate research assistant in DEPEND Group at Coordinate Science Laboratory (CSL) and pursuing PhD in computer science at University of Illinois, Urbana - Champaign. My research interests include design of fault tolerance/recovery methods for large-scale systems (HPC and Cloud) and data analytic frameworks to support resiliency studies. Currently, I am collaborating with NCSA, SNL, LANL, NERSC and Cray on "Holistic, Measurement-Driven Resilience" project. More details of my research activities in Depend group can be found at subgroup page - "Resiliency For eXtreme Scale Systems".
MS in Computer Science
Developed methods to automate the process of finding issues leading to virtual machine failures.
Researching in the topics of fault tolerance and reliability of large scale systems.
Completed research in the topics of In-memory hash joins for accelerators.
Teaching Assistant for Multicore System Programming Course. Took extra tutorial sessions for a batch of 30 Junior year students.
Worked as a Teaching Assistant for Artificial Intelligence Course. Took extra tutorial sessions for a batch of 60 undergraduate sophomore students.
Dr. Bingsheng He ,
NTU Singapore and
Mian Lu, IHPC, A* STAR, Singapore
In this work, we implement hash join algorithms on Xeon Phi and experimentally compare the performance of cache-conscious and cahce oblivious implementation of hash join on Xeon Phi. We also compare the different trends of hash joins on wide range of parameters on Intel Phi and Intel Xeon. More information and code snippets can be found at project web page .
In this work, we proposed a GPU based parallel simulated annealing algorithm for mirrored Traveling Tournament Problem (mTTP)and test the available instances on nvidia CUDA devices. We also introduced a new mTTP instance IPL-09 modelled on Indian Premier League - one of India's most popular sports league. Applying the proposed algorithm on this instance, we were able to reduce the total distance traveled by 30.20% or roughly 37,000 kilo-meters. The proposed algorithm converged faster to best known solutions with a significant speed up of 50-80x in terms of the number of solutions explored per second.
Guide: Dr. Rajesh Kanna
The aim of the project was to parallelize and evaluate the performance of image processing hypergraph algorithms such as image denoising and image representation on CUDA devices. Overall increase in performance was 10-20x times over CPU's.
This work was presented at ICS 2013 in Eugene, Oregon and HPDC 2013, NY, USA.