| Appendix G |
5-Day Workshop on
High Performance Computing
August 7th – 11th, 2006
at GIK Institute of Engineering Sciences & Technology, Swabi in collaboration with
University of Illinois at Chicago, USA.
.
Day 1:
The workshop started with the introductory speech by Dr. Abdullah Sadiq, Rector GIK, in which he iterated the importance of Parallel and Distributed Computing. After his welcome address he introduced the Guest speaker, Dr. Ashfaq Khokhar, who is well known for his research in Parallel and Distributed computing.
Dr. Khokhar stared the morning session with the discussion of some issues which arise in the parallel computing paradigm. He also talked about the Analysis of Numerical Computation on Parallel platforms. In his presentation he talked about some of the problems that arise when we analyze a problem using Parallel platforms. The instructor introduced various real life problems where conventional sequential programming is not efficient. Such applications either have large data sets or have complex time consuming computation. The instructor discussed the simulation of applications such as Finite element modeling, Molecular dynamics, Fourier transform and Climate modeling, using parallel programming. It was also discussed that how can parallel programming be used to solve various NP-hard problems e.g; Traveling Salesman, N-Queens, Bin packing, 0/1 knapsack, Graph partitioning etc.
The evening session was handled by one of the research students of Dr. Khokhar, Mr. Fadi J. Almsalha ( Ph.D. student, CS, UIC) in which he gave an introduction to Linux and Beowulf Cluster. Some basic Linux commands followed after the introduction.
Day 2:
Parallel computing techniques and Algorithm analysis was covered in the morning session. This included task decomposition and then generating a task dependency graph. The various tasks are mapped to processes and then processes are assigned to processors. The decomposition techniques include recursive decomposition, data decomposition, exploratory decomposition and hybrid decomposition.
The evening session was conducted by another research student of Dr. Khokhar, Mr. Waseem Ahmad ( Ph.D. Student, ECE, UIC). He covered configuration of Beowulf cluster.
Day 3:
In this session the instructor discussed various architectural models for HPC. He discussed that how can multiple computing platforms work in collaboration to yield high performance computing (HPC). State of the art HPC hardwares were discussed:
•Highly Parallel Machines
•SMPs
•Clusters
•Grid The instructor also presented various memory models:
•Shared Memory
•• UMA (Uniform Memory Access) vs NUMA (Non-Uniforms Memory Access)
•Distributed Memory
•• Local Address Space vs Global Address Space
In the evening session, the participants were given an opportunity to have hand-on experience with the Message Passing Interface (MPI), which is an industry standard for writing message passing programs on HPC platforms. MPI is a specification for the developers and users of message passing libraries. Interface specifications have been defined for C/C++ and Fortran programs.
Day 4:
In this session the instructor gave an Introduction to Shared Memory Programming and OpenMP Standard. OpenMP is an open specification for Multi-Processing. And is a Standard API for defining multi-threaded shared-memory programs. In the evening session we did parallel programming using OpenMP.
Day 5:
This session included Advance parallel computing techniques. The participants were asked to demonstrate various research problems and discuss the design using HPC. I discussed the clustering problem. I also told them about the domains in which they are used in the real world such as Internet Search Engines, Optimal Placement of Components in VLSI design etc. The participants were given an introduction of the problem and some of the simulation results were also discussed.
The evening session was a wrap up session. The topics covered and the issues discussed throughout the workshop were summarized. The workshop ended with certificate distributing ceremony.
Conclusion:
The workshop proved to be very informative as a whole. Different programming techniques and problem solving skills were discussed as well. The hands on training proved to be really effective in understanding the core concept used in the underlying technology. The results of the workshop can very well be utilized in the following manner:
1. Possibilities of research collaboration with Dr. Khokhar should be explored.
2. Offering of Student FYPs in the Parallel Computing domain.
3. Establishment of a Cluster for parallel processing at FAST.
If these suggestions are followed, then the course can be best utilized in context of improving the research activities of the students and faculty of FAST. A very good thing about the implementation of these steps is that no extra Hardware/Software is required for the research group. The existing Hardware/Software will suffice if some research work were to be started. The research could very well lead to a more advanced research of GRID Technologies which is used all over the world for parallel processing across geographical boundaries.
