Advisor Prof. William D. Gropp, Department of Computer Science,
University of Illinois,Urbana- Champaign, USA.
Environment C++, POSIX, MPI
Description The objective of the project is to improve the performance of Dendro, a Geometric Multigrid package to solve Finite element Problems. By performance, it means both single node performance using POSIX threads, as well as parallel performance using MPI. Also, developing a Performance-Model to evaluate the impact of memory references which affects both the single node and parallel performance.
Advisor Prof. Indranil Gupta, Department of Computer Science,
University of Illinois, Urbana- Champaign, USA.
Environment Hadoop (Open-source Map-Reduce implementation by Yahoo), Java
Description The objective of the project is to identify the different sources of noise (we call it Jitter) for applications running on Hadoop. We developed a scheduling policy to mitigate the jitter-effect. The jitter is the one of the main impediment towards scalability. So, by implementing the jitter-aware algorithm on top of the default FIFO scheduler of Hadoop, better performance predictability and scalability can be achieved.
Advisor Dr. Soumyendu Raha,
Supercomputer Education and Research Centre, IISc, Bangalore, India.
Environment C++, POSIX, LAPACK, BLAS, MPI, Simics, CACTI, eCACTI
Description The objective of the project was to exploit the availability of several processing element on same chip, a re-look into traditional linear algebra operations and algorithms by utilizing the on-chip speed and availability of hardware level multithreading. The data distribution has been done by tuning the algorithms so that a particular thread will operate on a specific portion of the matrix. The effect of cache hierarchy has been considered to improve the performance and tune the algorithm to ensure minimum cache miss. We have simulated Niagara with its characteristic cache-hierarchy model using Simics and CACTI and eCACTI is being used to model the cache power characteristic.
Advisor Prof. Pradip Datta, Mechanical Engineering, IISc, Bangalore, India
Environment C++, Fortran90, MATLAB 7.1
Description Electromagnetic induction principle is used for calculating the heat generation for various manufacturing processes. Applying Maxwell equations and Faradays law of electromagnetic induction the induced e.m.f. can be calculated. Due to axi-symmetry, 3D model can be represented by 2-D model to predict the heating of the workpiece due to applied time varying magnetic field.