Research Interests: High Performance Computing, Optimization algorithms, Performance Analysis & Optimization, Energy Modeling.

Dissertation Area: Scalable optimization algorithms for problems in scientific computing.

Advisors: Dr.G(Kumar)Mahinthakumar (Committee Chair), Dr. Joe DeCarolis.

A case study currently being investigated is the contaminant source characterization problem in a water distribution system(WDS). This entails finding the contaminant source location(s) as well as the release pattern(s) over time.

I'm concurrently involved in an Open Source Energy Systems Modeling and Optimization project.

Recent work done during a research fellowship at EPFL,Switzerland involved designing a multi-objective optimization algorithm to perform model fitting for Neuroscience models.

Master's thesis: Cyberinfrastructure for Contamination Source Characterization in Water Distribution systems.

Advisors: Dr.G(Kumar)Mahinthakumar (Committee Chair), Dr.Xiaosong Ma (Committee Co-Chair), Dr.Ranji Ranjithan, Dr.Frank Mueller

Book chapters

• Kumar J., Sreepathi, S., Ranjithan, S., Brill, E. D., Liu, Li, Mahinthakumar, G., (2009), “Cyberinfrastructure for contamination threat management in water distribution systems”, In Darema, F., Douglas, C., (eds.) , Dynamic Data Diven Analysis Systems, Sprin-Verlag. (to be published in 2010)

Publications

Cyberinfrastructure for Threat Management in Water Distribution networks
- National Science Foundation DDDAS (Dynamic Data-Driven Application Systems) Project

(Read a outline of the activities in the Cyberinfrastructure section)

Developing a framework for parallel bio-inspired optimization algorithms.

• Investigation of Swarm Intelligence algorithms for solving the contaminant source identification problem.

Porting simulation code to Blue Gene/L and Cray XT4. 

Grid enabled workflow for solving water quality simulations to detect contamination in urban water distribution networks.

• Coarse grain parallelization of water quality simulation code(EPANET)

• Optimization methods using Evolutionary Computing techniques and concepts from Graph Theory.

• Workflow encompassing interactions between the optimization methods and parallelized water quality simulation code.

• Framework for deploying on large scale distributed systems like Teragrid and SURAgrid.

• Performance analysis and optimization.
  

Links:

Secure-Water Project Website

Secure-Water Wiki

• Build an open source energy system model.The initial version will consist of a technology rich, partial equilibrium model. As development proceeds, we will account for macroeconomic effects.
• Design the model in a way that makes uncertainty analysis easy to perform. We plan to automate sensitivity analysis, Monte Carlo simulation, multi-stage stochastic optimization, and search techniques to find near-optimal alternative solutions.

Besides model design and development, my focus would be on nonlinear optimization algorithms and search techniques for generating alternative solutions.

TEMOA Project Website

Performance Analysis and Optimization of Groundwater application
- Dept. of Energy SciDAC Project (Scientific Discovery through Advanced Computing)

Performance analysis and optimization of PFLOTRAN, esp on Cray XT4 (Second fastest supercomputer upon its debut in June '07) at Oak Ridge National Laboratory. This work is part of an ongoing SciDAC PERI project.

Preliminary results indicate an improvement of 18.77% over baseline results on Cray XT4 while running on 4096 cores.

Links: 

Recent Results

Other Projects

Performance Analysis and Tool Evaluation
- Dept. of Energy SciDAC Project (Scientific Discovery through Advanced Computing)

Analyzing performance of High Performance Computing Applications including performance tool evaluation at several Supercomputing facilities (Oak Ridge National Laboratory, Teragrid etc.)

Links: Visit my earlier website for results.

• Performance Analysis of a nuclear fusion application Gyro - Results .
  

• Performance analysis of GenASiS (GENeral Astrophysics SImulation System for the Terascale Supernova Initiative, Oak Ridge National Laboratory) - Results.

• Activities included manual analyis, Profiling and Tracing using parallel performance analysis tools.

• Benchmarking, Scalability analysis and cross platform performance comparison.

• The following tools were used for the performance tool evaluation.

  • TAU : Tuning and Analysis Utilities, SvPablo: A Toolkit for Performance Analysis of Parallel Systems, PerfSuite from NCSA, HPC Toolkit from Rice University, mpiP from LLNL, Other tools from PERC (Performance Evaluation Research Center).
    

Research Clusters

Cygnus and Neptune are two Opteron Clusters in the Department of Civil Engineering at North Carolina State University and are our primary compute resources. Stargate is a new IBM cluster that is being setup to assess high speed networking(10 GigE, Infiniband) technologies. We have a total of 22+88+40 compute processing elements.

• I have planned, configured and deployed the clusters.

• Conducted extensive benchmarking using High performance Linpack(HPL) and Matlab Distributed Computing Toolbox among others.

• Achieved a peak performance of 76 GFLOPS with HPL(Linpack) while solving a solving a 70K x 70K dense matrix, corresponding to 78% of the theoretical peak on Neptune.

Presentations

• Performance analysis tools: A guest lecture (ppt) was given on parallel performance analysis and profiling tools for graduate level Parallel Computing course at NCSU.

• Blue Matter (Blue Gene): Class Presentation(ppt, pdf) on the architecture of Blue Gene,the world's fastest supercomputer and Blue Matter, the application framework for molecular simulation on Blue Gene was given as part of PALM research group seminar series at NCSU. 

• Secure Water

• Energy Modeling

  Other Activities

• Parallel Applications and Middleware (PALM)

• Evolutionary Computation (Student Seminars - Fall '06)