coe501

First 1-4 weeks of semester: Students are expected to independently carry out literature Survey on their selected topics, compile information of relevant papers in a progress report PR-1, have an action plan of what they are going to contribution and more specifically on their project implementation, and prepare for a presentation to expose all the above aspects to the class and instructor. During this period, the student is expected to interact with the instructor regarding the major issues of his task while being prepared for his topic and its details. A 15-minute presentation is to be given in week 5 in front of the class and instructor. This part will be rated 25% of project for both presentation and timely submission of PR-1.
From week 5 to week 10, Students are expected to carry out the implementation part of their project by taking into account: (1) his presented or revised action plan, and (2) the instructor observations formulated during or after his presentation. During this period, the student is expected to interact with the instructor regarding the major design issues, difficulties in addressing some parts, resource problems, or any kind of problem that require specific attention to avoid negative impact on the project. Prepare report PR-2 after adding the implementation aspects presented above to P-1. A 15-minute presentation involving success and failure design and implementation aspects is to be given in week 11 in front of the class and instructor. This part will be rated 35% of project for both presentation and timely submission of PR-2.
From week 11 to week 14, Students are expected to carry out: (1) debugging and testing of their implementation and refer to instructor in case of problems, (2) carry out performance evaluation of the project with data collection, (3) revise PR-1 by including the performance evaluation and detailed analysis of collected results, (4) prepare the final project presentation. During this period, the student is expected to interact with the instructor regarding the debugging issues and performance interpretation. A presentation involving overall project will be delivered by the team (or student) before the last day of classes (week 15). The emphasis should be here on the method and results. This part will be rated 40% of project for both presentation and timely submission of PR-3 which is the final report. Here the student is to submit a zipped folder including all of original word reports (not pdf), well documented source code, evaluation data, and reference papers. A demo to the instructor will also be required.

IBM Cluster Compiling Systems: A HPC cluster is provided with FORTRAN 77 (Portland Group compiler pgf77), Fortran 90 (pgf90), C (pgcc), and C++ (pgCC) compilers in addition to either of the Intel or / and Portland Group suites of optimizing compilers (faster code than that GNU compilers).Generally these different produce Linux executable for each of the Portland Group and Intel compilers. There are also compiler optimization options to produce more optimized executable codes.

Parallelization over one Multi-Core Node Acting as a Shared-Memory Multiprocessor: Users can automatically optimize single-node sequential programs for shared-memory parallel execution using the Portland Group -Mconcur or Intel -parallel compiler option (pgcc -O2 -Mconcur sample.c or icc -O2 -parallel sample.c). Both the Fortran and C/C++ compilers understand the OpenMP set of directives, which give the programmer a finer control over the parallelization. The -mp (Portland Group) and -openmp (Intel) compiler options activate translation of source-level OpenMP directives and pragmas. This allows compiling for OpenMP threaded execution, runs the executable using a number of threads on 1 node (multi-core), each thread on a separate core.

Parallelization over Distributed-Memory System using the Message Passing Interface (MPI): The system uses the MPICH implementation of the Message Passing Interface (MPI), generally optimized for the high-speed Infiniband interconnect. MPI is a standard library for performing parallel processing using a distributed-memory model. Each program file using MPI must. Near the beginning of each C or Fortran source file we must include the MPI header file. Generally an MPI wrapper scripts (Portland Group or Intel compilers) is loaded prior to executing the compilation command. The MPI compilers take the same options as the compiler they wrap. A command “mpiexec” (the -pernode for command line options) spawns one MPI process per CPU in a batch job. The -pernode option requests that one MPI process be spawned per node. These options are intended to be used for codes which mix MPI message passing with some form of shared memory programming model, such as OpenMP or POSIX threads. Finally a GNU debugger gdb is recommended for interactive or post analysis of sequential programs.