240 likes | 381 Views
CS6035 Parallel/Distributed Processing II:. An Efficient Static Assignment Parallelization Scheme for Algebraic Fractals. By: Chris MacPhee Supervisor: Dr. Bhavsar. Outline:. Introduction Computational Characteristics Serial Program Parallelization Experimental Results IBM SP
E N D
CS6035 Parallel/Distributed Processing II: An Efficient Static Assignment Parallelization Scheme for Algebraic Fractals By: Chris MacPhee Supervisor: Dr. Bhavsar
Outline: • Introduction • Computational Characteristics • Serial Program • Parallelization • Experimental Results • IBM SP • SGI Onyx • Conclusion
Introduction What are fractals? • Possess non-Euclidian geometry (“formless”) • Self-similar (same type of structure at all scales) • “Clouds are not spheres, mountains are not cones, coastlines are not circles, and bark is not smooth, nor does lightning travel in a straight line.” • - Benoit Mandelbrot, 1983
Introduction Examples of fractals images From: Fractal Gallery http://projekt.pinknet.cz/fractal/
Computational Characteristics The Mandelbrot set • zz2 + c, where z, c • z0 is a constant • c varies • z is iterated until either: • z diverges beyond a preset limit • the maximum number of iterations • is reached
Computational Characteristics The Mandelbrot set (cont) zz2 + c
Computational Characteristics The generalized function zz2 + c, where z, c zz + c, where z, c
Computational Characteristics The generalized function (cont) zz10 + c
Computational Characteristics The generalized function (cont) zz100000 + c
Computational Characteristics Distribution of iterations
Parallelization Two architectures • Shared memory programming • Run on SMP machines (e.g. Sun & SGI) • Uses OpenMP • Message passing programming • Run on distributed memory machines • (e.g. Compaq & IBM) • Uses Message Passing Interface (MPI)
Parallelization Three work assignments Static work assignment Dynamic work assignment New static work assignment
Parallelization Static work assignment Divide column groups evenly between processors Master 1-256 257-512 513-768 769-1024
Parallelization Dynamic work assignment Farm work to the slaves in work sizes of 64 columns Master (in queue: 513-1024) 449-512 321-384 385-448 257-320
Parallelization New static work assignment Divide workload evenly over processors Master 1-339 370-513 514-657 658-1024
Experimental Results Two machines • Symphony (University of New Brunswick) • IBM SP • 16 375 MHz processors • 4 GB of RAM • Distributed memory architecture • Herzberg (Memorial University of Newfoundland) • SGI Onyx • 28 400 MHz processors • 14 GB of RAM • Shared memory architecture
Experimental Results IBM SP Timings Computing time for each slave processor for = 2
Experimental Results IBM SP Timings Computing time for each slave processor for = 10
Experimental Results IBM SP Timings Computing time for each slave processor for = 100000
Experimental Results SGI Onyx Timings Computing time for each slave processor for = 2
Experimental Results SGI Onyx Timings Computing time for each slave processor for = 10
Experimental Results SGI Onyx Timings Computing time for each slave processor for = 100000
Summary Summary • The computational characteristics of fractal images have been analyzed. • A static assignment method for efficient parallel processing has been developed. • The static assignment method becomes more efficient as increases.
References [1] H. O. Peitgen and P. Richter, The Beauty of Fractals, Springer-Verlag, Berlin, 1996. [2] U. G. Gujar and V. C. Bhavsar, "Fractals from z z a + c in the Complex z- plane", Comp. and Graph., 16(1), pp. 45-49, 1992. [3] S. V. Dhurandhar, V. C. Bhavsar, and U. G. Gujar, "Analysis of z-plane fractal images from z z a + c for a < 0", Comp. and Graph., 17(1), pp. 89-94, 1993. [4] V. C. Bhavsar, U. G. Gujar, N. Vangala, "Vectorization of generation of fractals from z z a + c on IBM 3090 / 180VF", Comp. and Graph., 17(2), pp. 169-174, 1993. [5] E. Aubanel, "Parallel Programming with Generalized Fractals," Faculty of Computer Science, University of New Brunswick, February 2002, http://www.cs.unb.ca/profs/aubanel/aubanel_fractals.html. [6] B. Wilkinson and M. Allen, Parallel Programming, Prentice Hall, Upper Saddle River, 1999.