Transcript poster
Independent Study of Parallel Programming Languages An Independent Study By: Haris Ribic, Computer Science - Theoretical Independent Study Advisor: Professor Daniel Bennett WHY THIS STUDY? “In 2007, the degree of parallelism for personal computing in current desktop systems such as Linux and Windows Vista is nil, which either indicates the impossibility of the task or the inadequacy of our creativity.” – Gordon Bell in Communications of the ACM Everybody has seen these advertisements but what do they mean? MPI-Message Passing Interface UPC-Unified Parallel C OpenMP Standard based library that allows many computers to communicate with one another. An extension of the C programming language designed for high-performance computing on large-scale parallel machines An application programming interface that supports multiplatform shared memory and multiprocessing programming. ... moreover we have seen an increase in DUAL-CORE processors but still what does it all mean? Simple schematics of a DUAL-CORE processor build by Intel and AMD. Charm++ Preliminary results using Monte Carlo Method for calculating Pi (π) using MPI and OpenMP languages on a Computer Cluster and Symmetric Multiprocessing Computer. 2 Simple schematics of a computer cluster like the one used by Computer Science Department. 0 Monte Carlo Method - Generate random numbers x and y ranging from 0 to 1 - Count the number of hits inside the circle by x^2 + y^2 < 1 - Probability (Hit) = Surface Circle/Surface Square - Π = number of hits in circle / total hits - Generating more numbers increases accuracy of the number π - However more numbers slow down the computer 2 MPI OpenMP MPI OpenMP 120000 100000 9000 8000 7000 Process 1 6000 5000 4000 3000 2000 1000 0 Program Process 2 … Process K 5E+07 8E+07 1E+08 2E+08 2E+08 3E+08 5E+08 8E+08 Sequential - Ability to parallelize programs - CPU uses less energy and delivers more performance - Better system responsiveness and multi-tasking capability - Ability to parallelize programs - More cost-effective supercomputer - Used in scientific research “For example you could have your Internet browser open along with a virus scanner running in the background, while using Media Player to stream your favorite radio station and the dual-core processor will handle the multiple tasks without the decrease of performance and efficiency.” www.intel.com/core2duo A computer cluster is a group of loosely coupled computers that work together closely so that in many respects they can be viewed as though hey are a single computer. The components of a cluster are commonly connected to each other through fast local area networks. 60000 40000 20000 5000000 1E+09 2 PC's 4 PC's 10000000 25000000 50000000 100000000 1000000000 Number of Trials Number of Trials Why Computer Cluster? 80000 0 3E+07 Why DUAL-CORE? Timing Timing 12000 11000 10000 System BUS Parallel object-oriented programming language based on C++. Sequential 6 PC's - Efficiency increases when using more nodes - Language independent, could use C++, FORTRAN - Difficult to use Master Node Nodes 1 0 2 … K MPI_Init(&argc, &argv) MPI_Comm_rank(MPI_COMM_WORLD, &my_rank) MPI_Comm_size(MPI_COMM_WORLD, &p) MPI_Barrier(MPI_COMM_WORLD) Master Node 0 MPI_Bcast(&totalDarts, 1, MPI_INT, src, MPI_COMM_WORLD) Each node performs calculation MPI_Gather(&time, 1, MPI_INT, aryTime, 1, MPI_INT, src, MPI_COMM_WORLD) Thread Thread MPI_Reduce(&hits, &allHits, 1, MPI_INT, MPI_SUM, src, MPI_COMM_WORLD) Print results CPU 1 quad - Easy to write - Depends of operating system scheduling #pragma omp parallel private(i, trdID) shared(rndAry, hits, darts) trdID = omp_get_thread_num() srand(rndAry[trdID]) #pragma omp for reduction(+:hits) for (i = 0; i < darts; i++) x_value = drand48() y_value = drand48() if ( ((x_value*x_value)+(y_value*y_value)) <= 1 ) hits = hits + 1 Process Get number of darts single HT CPU 2