J. Demmel and S. L. Smith. Parallelizing a global atmospheric chemical tracer model. In Proceedings of the Scalable High Performance Computing Conf., pages 718--725, Knoxville, TN., May 1994.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....systems. 2.1. Irregular Communication In irregular many to many communication, some nodes concurrently read from and write to some other nodes and the size of the messages transmitted varies from one to the other. Such communications arise in scientific computations in environmental sciences, [16], in high level vision problems [14, 13] and in network simulations [11] In [11] irregular problems arising in scientific and industrial applications have been classified into loosely synchronous, asynchronous, embarrassingly parallel, and meta problem categories; these applications result in ....

J. Demmel and S. Smith, "Parallelizing a Global Atmospheric Chemical Tracer Model," Symposium on High Performance Computing and Communications, May 1994.

....few coarse grained tasks and execute on heterogeneous systems formed by a few machines [1, 15, 17] These machines can be MIMD MPPs, visualization engines, SIMD MPPs, workstations, and or vector computers. For many of these applications, both heterogeneity and parallelism in the code are exploited [4, 10, 13, 14]. For instance, in a molecular structure determination application which computes the global minimum of a potential energy function [14] 1. Supported by a scholarship from CAPES and UFRJ (Brazil) 2. Supported in part by NSF contract number ASC 9301788. TR UCSD CS96 484 2 an MPP is used ....

J. Demmel and S. Smith, "Parallelizing a Global Atmospheric Chemical Tracer Model," Proceedings of the Scalable High-Performance Computing Conference, pp. 718-725, August 1994.

....For many real world applications, we need all of: processors capable of high sustained floating point performance, networks with bandwidth that scales with the number of processors, parallel file I O, and low overhead communication. One example is the AMES UCLA chemical tracer model (GATOR) [DeSm94]; it models atmospheric chemistry in the Los Angeles Basin and has been used for detailed air pollution studies. Amodel has been developed of GATOR s execution time as a function of various input parameters (grid resolution, number of chemical species) and system parameters (CPU floating point ....

J. Demmel and S. Smith, "Parallelizing a global atmospheric chemical tracer model",Symposium on High Performance Computing and Communications, May 1994.

....and runs on multiple platforms. 1. Introduction Judicious choice of data layouts to balance computation and minimize communication is critical to the performance of parallel programs. Even for regular computational kernels, determining the optimal layout may require detailed performance models [10] or experimentation [6] For large parallel programs composed of irregular and regular kernels, experimenting with several implementations is not feasible. In this paper, we apply a simple latency bandwidthmodel to the design of a cell simulation application, which has a regular phase that ....

J. Demmel and S. Smith. Parallelizing a global atmospheric chemical tracer model. In IEEE Conference for Scalable High Performance Computation, May 1994.

....and its parallel performance from the start of its parallel design. This involved developing a detailed performance model to predict the number of wall clock seconds needed to simulate one day of climate as a function of GATOR ACTM parameters, architecture parameters, data layout and algorithm [3]. During the design phase of the parallel implementation, this performance model proved to be an invaluable tool in identifying potential bottlenecks, eliminating design alternatives, and giving feedback to earth scientists as to what simulations they could expect to run. In this paper we present ....

....of time required by the ODE solver routine for each cell. 4 3 Parallelization of GATOR ACTM During the design of the parallel implementation of GATOR ACTM, we evaluated several static partitioning strategies and their performance behavior for several different parallel transport frameworks [3]. This section describes the design choices we made, based on those that held the most promise for meeting the goals of an efficient and portable parallel code. One of the first decisions to be made in any parallel implementation is the type of data layout to use. In the atmospheric chemistry ....

J. Demmel and S. L. Smith. Parallelizing a global atmospheric chemical tracer model. In Proceedings of the Scalable High Performance Computing Conf., pages 718--725, Knoxville, TN., May 1994.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC