19 citations found. Retrieving documents...
Rolf Hempel and Hubert Ritzdorf, The GMD Communication Library for Grid-Oriented Problems, Gesellschaft fur Mathematik und Datenverarbeiung, mbH, Institute F1/T, Postfach 1240, 5205 St. Augustin, Germany, Oct. 1991. Department of Computer Sciences & Center for Relativity ffl University of Texas at Austin Austin, TX 78712-1081

 Home/Search   Document Not in Database   Summary   Related Articles   Check  

This paper is cited in the following contexts:
MPI-RGL: a Regular Grid Library for MPI - Version Clive Baillie   (Correct)

....a simple interface to the application specialist who does not wish or need to know the details of the underlying message passing library. To date most of these usability layers have focussed on regular grid problems, for example: NNT [5] BlockComm [6] RSL [7] and the GMD Communications Library [8]. Thus we have abstracted a minimal set of functions from all of these which are sufficient to implement any regular grid based code. There are five basic functions. The first domain decomposes the grids in up to three dimensions, and the second declares arrays for the local piece of the grid in ....

....specification. All of the versions 0.x should be considered preliminary, development, beta releases; the first complete edition will be called version 1.0. 4 2 Background There are at least four usability layers in use today: NNT [5] BlockComm [6] RSL [7] and the GMD Communications Library [8]. They all help in coding regular grid problems on parallel computers. We shall briefly review each one, pointing out the ideas we absorbed into MPI RGL and those we did not. NNT, the Nearest Neighbor Tool, is a library which has been developed at the NOAA (National Oceanic and Atmospheric ....

Rolf Hempel, Hubert Ritzdorf, "The GMD Communications Library for Grid-Oriented Problems," GMD, Germany (1991).


MPI-RGL: a Regular Grid Library for MPI - Version Clive Baillie   (Correct)

....to the application specialist who does not wish or need to know the details of the underlying message passing library. To date most of these usability layers have focussed on regular grid problems, for example: NNT [HaHeRo94] BlockComm [Gro94] RSL [Mic94] and the GMD Communications Library [HeRi94]. Thus we have abstracted a minimal set of functions from all of these which are sufficient to implement any regular grid based code. There are five basic functions. The first domain decomposes the grids in up to three dimensions, and the second calculates the local start and end indices for ....

....MPI forum and hope for some concensus on what it should look like. So far we have only partially implemented MPI RGL and coded the 2D Laplace solver. 2 Background There are four main usability layers in use today: NNT [HaHeRo94] BlockComm [Gro94] RSL [Mic94] and the GMDCommunications Library [HeRi94]. They all help in coding regular grid problems on parallel computers. We shall briefly review each one, pointing out the ideas we absorbed into MPI RGL and those we did not. NNT, the Nearest Neighbor Tool, is a library which has been developed at the NOAA (National Oceanic and Atmospheric ....

Rolf Hempel, Hubert Ritzdorf ; "The GMD Communications Library for Grid-Oriented Problems" GMD, Germany, 1991


The Same-Source Parallel MM5 - Michalakes (1998)   (1 citation)  (Correct)

....better scaling because locality is explicitely enforced. Much of the painful low level detail originally associated with message passing programming domain decomposition, message passing, distributed I O, and load balancing has been efficiently encapsulated in application specific libraries (Hempel and Ritzdorf, 1991; Kohn and Baden, 1996; Michalakes, 1997c; Parashar and Browne, 1995; Rodriguez et al., 1995) These approaches still require modification to the code for iteration over local data, global and local index translation, and distributed I O. If one is able to design a new model or undertake a major ....

Hempel, R., and H. Ritzdorf, 1991: The GMD Communications Library for Grid-oriented Problems, Tech. Rep.


Performance Portability for Coupled Atmosphere-Ocean General.. - Worley   (Correct)

....although partly because it was unfamiliar. Now, however, much of the painful low level detail originally associated with message passing programming domain decomposition, message passing, distributed I O, and load balancing can be encapsulated in libraries, as many groups have demonstrated [19, 36, 31, 27, 37]. These approaches still require significant modification to the code for handling loops over local data, global and local index translation, and distributed I O. Additionally, the ability to encapsulate does not guarantee that the encapsulation provides the desired performance portability. As ....

R. Hempel and H. Ritzdorf. The GMD Communications Library for Grid-oriented Problems. Technical Report GMD-0589, German National Research Center for Information Technology, 1991.


MM90: A Scalable Parallel Implementation of the Penn State/NCAR .. - Michalakes (1997)   (2 citations)  (Correct)

....multiple nests. It uses vendor specific message passing layers when available (NX on the Paragon, MPL on the IBM SP2) and MPI on other platforms. RSL was developed at Argonne National Laboratory in conjunction with MPMM and MM90. It is similar to packages developed by a number of other groups [4][5] 10] 11] MPMM and its follow on Fortran90 implementation, MM90, are functionally equivalent in most respects to the 1995 release of MM5 (MM5v1) The parallel model is strictly nonhydrostatic, although a separate hydrostatic version of MPPM has been implemented by Y. Kim at Iowa State. Since ....

R. Hempel and H. Ritzdorf, The GMD Communications Library for Gridoriented Problems, Tech. Rep. GMD-0589, German National Research Center for Information Technology, 1991.


The Distributed-Memory Parallel Option to MM5 - John Michalakes, Jim Tuccillo   (Correct)

....underlying parallelization library to use. Currently, the Argonnedeveloped RSL runtime system library [3] 6] is provided (the library and sources are public domain) but this has been designed to be interchangeable, allowing the use of other existing or future communication packages (e.g. 8] 7][2]) This added flexibility will help assure that the same source MM5 code remains adaptable to future architectures and programming models. 1 See http: www.mcs.anl.gov Projects FLIC. 4 Conclusion We have described an effort that will expand the set of architectures that will run the NCAR ....

R. Hempel and H. Ritzdorf, The GMD Communications Library for Grid-oriented Problems, Tech. Rep. GMD-0589, German National Research Center for Information Technology, 1991.


Reference Manual for RSL: A Runtime System Library for Parallel .. - Michalakes   (Correct)

....at Argonne National Laboratory that is tailored to regular grid atmospheric models with nesting. It was used with Fortran90 to parallelize a wellknown and widely used regional weather model, the Penn State NCAR Mesoscale Model MM5 [1] It is similar to efforts of a number of other groups: Comlib [2], LPARX [3] and NNT SMS [5] in that it provides high level, efficient mechanisms for performing data parallel computations over multiple interelated grids. Like the LAPRX software abstractions, RSL is able to support dynamic load balancing, efficient inter grid communication, irregularly shaped ....

R. Hempel and H. Ritzdorf. The GMD communications library for grid-oriented problems. Technical Report GMD-0589, German National Research Center for Information Technology, 1991.


FLIC: A Translator for Same-Source Parallel Implementation of.. - Michalakes (1997)   (2 citations)  (Correct)

....boundary conditions specified implicitly as loop ranges do not hold, because the loop ranges are global whereas the loop variable is used as a local index. All of these problems have been successfully addressed using run time systems and parallel libraries. A number of available packages [2, 3, 5, 6] provide the necessary looping and index translation information to the program at run time. However, looping statements and index expressions must be modified from their original form to use this run time information, introducing enough parallel artifact to hinder a same source software approach. ....

R. Hempel and H. Ritzdorf, The GMD communications library for grid-oriented problems, Tech. Rep. GMD-0589, German National Research Center for Information Technology, 1991.


Performance And Portability In Parallel Computing: .. - Bernardo.. (1995)   (Correct)

....I O, remote data reference (including exchanges) and global operations. Express [6] combined low level message passing and I O routines with higher level calls for data decomposition, global operations, and two dimensional I O. A high level library for grid oriented problems was proposed in [7]. It allows one to define a block structured grid with boundaries between blocks, or overlap regions. The data in the overlap regions define the data that needs to be communicated through exchanges between processors. The Nearest Neighbor Tool (NNT) is a library that isolates the user from all ....

R. Hempel, and H. Ritzdorf, "The GMD Communications Library for GridOriented Problems," Technical Report 589, GMD, Sankt Augustin, Germany, 1991.


A Runtime System Library for Parallel Finite Difference Models.. - Michalakes (1995)   (1 citation)  (Correct)

....Section 5 walks through the parallelization of a simple relaxation code using RSL. Section 6 discusses the implementation of a MM5 using RSL. Section 7 presents conclusions and directions for future work. 2 Related Work RSL is similar to efforts of a number of other groups, particularly Comlib [5], LAPRX [6] and NNT SMS [9] in that it provides high level, efficient mechanisms for performing dataparallel computations over multiple interrelated grids. Like the LAPRX software abstractions, RSL is able to support dynamic load balancing, efficient intergrid communication, irregularly shaped ....

R. Hempel and H. Ritzdorf, The GMD communications library for grid-oriented problems, Tech. Rep. GMD-0589, German National Research Center for Information Technology, 1991.


Parallelizing Operational Weather Forecast Models For.. - Bernardo Rodriguez (1995)   (Correct)

....of changes introduced by NNT be minimal, but the resulting loop structures and computation lines must also be identical to the original versions, as shown in Section 3. Several libraries have been proposed to parallelize weather prediction models. Among them, the GMD Grid Communication Library [4] and the Argonne National Laboratory s RSL [8] seem to be the most promising [2] 9] Other efforts to parallelize weather prediction models have been based on highlevel libraries [1] 7] The GMD library is of more general scope than NNT, since it allows for nonregular domains. The cost of this ....

R. Hempel, and H. Ritzdorf. The GMD communications library for grid-oriented problems. Tech. Rep. 589, GMD, Sankt Augustin, Germany, 1991.


Rsl: A Parallel Runtime System Library For Regional Atmospheric .. - Michalakes   (2 citations)  (Correct)

....and portability of the code. At first, few tools existed beyond low level message passing libraries. Over time, however, many groups have developed libraries that roll in other necessary functionality: library level support for data domain decomposition and computation over distributed domains [1][9][18] mesh refinement, and load balancing [10] 17] The Runtime System Library [12] was developed at Argonne the course of a research effort to address load imbalance and nesting in the development of regional weather models. In addition to providing higher level communication constructs, stencil ....

R. Hempel and H. Ritzdorf, The GMD Communications Library for Gridoriented Problems, Tech. Rep. GMD-0589, German National Research Center for Information Technology, 1991.


Interoperability of Data Parallel Runtime Libraries with.. - Edjlali, Sussman, Saltz (1997)   (8 citations)  (Correct)

.... not exist in widely available data parallel languages, such as High Performance Fortran (HPF) 14] Examples of such data parallel runtime libraries include AMR and P [17, 20] for adaptive grid applications, Grids [9] for structured and unstructured grid problems, Multiblock Parti [1] and GMD [11] for multigrid and multiblock codes, LPARX [15] for codes using adaptive finite difference methods, Chaos [7] and PILAR [16] for unstructured grid problems, GA [18] for computational chemistry and Aztec [12] PETSc [10] ScaLAPACK [5, 6] and Syisda [3] for linear algebra operations. Although this ....

Rolf Hempel and Hubert Ritzdorf. The GMD communications library for grid-oriented problems. Technical Report 589, GMD, November 1991.


Data Abstraction Techniques for Multilevel Algorithms - Rüde (1992)   (Correct)

....The situation outlined in Section 1 is unsatisfactory. These difficulties are realized by many developers. Some problems can be alleviated by classical modularization techniques, e.g. by using parallel base libraries to make a code portable to various parallel machines, see Hempel and Ritzdorf [6]. However, the general problems caused by the inherent complexity of multigrid in combination with insufficient software design techniques remains. In the following we will try to analyze some of the difficulties. 2.1 Data organization In current programming practice the storage layout of the ....

R. Hempel and H. Ritzdorf. The GMD communications library for grid-- oriented problems. Arbeitspapiere der GMD 589, GMD, St. Augustin, 1991.


Data Structures for Multilevel Adaptive Methods and Iterative.. - Rüde (1992)   (2 citations)  (Correct)

....the data structure for each patch can be comparatively simple. If the patches are non overlapping, they can be considered as macro elements. Interface conditions describe their interaction with neighboring patches. Techniques related to this idea are block structured grids (see Hempel and Ritzdorf [28]) domain decomposition (see Quateroni [43] and the references therein) or the substructuring technique. These techniques describe solution algorithms suitable for piecewise uniform grids. For the overall triangulation to be conforming, the nodes along the patch boundaries must be conforming ....

R. Hempel and H. Ritzdorf, The GMD communications library for grid--oriented problems, Arbeitspapiere der GMD 589, GMD, St. Augustin, 1991.


The ANL/GMD Macros (PARMACS) in FORTRAN for Portable Parallel.. - Hempel (1991)   Self-citation (Hempel)   (Correct)

....such automatic systems are still in their infancy. There the usual situation is that in porting an application program from one machine to the other a large fraction of the code has to be rewritten. There have been solutions to the portability problem at least for specific classes of applications [4, 5, 6], but these do not help in the general case. Other approaches give satisfactory results if the grain size of the parallelism is not too fine. One example is the portable parallel language STRAND88 [3] which has a foreign language interface for including FORTRAN or C subroutines into a STRAND ....

Hempel, R., Ritzdorf, H.: The GMD Communications Library for Grid--Oriented Problems. GMD Arbeitspapier No. 589.


LiSS-PDE - Canu, Lonsdale, Ritzdorf (1994)   Self-citation (Ritzdorf)   (Correct)

No context found.

Hempel,R.; Ritzdorf,H.: The GMD Communications Library for Grid--Oriented Problems. -- User's Reference Manual --


LiSS-PDE - Canu, Lonsdale, Ritzdorf (1994)   Self-citation (Ritzdorf)   (Correct)

No context found.

Hempel,R.; Ritzdorf,H.: The GMD Communications Library for Grid--Oriented Problems. -- User's Reference Manual --


A Survey of Existing Support DRAFT Scalable Distributed.. - Parashar And Browne   (Correct)

No context found.

Rolf Hempel and Hubert Ritzdorf, The GMD Communication Library for Grid-Oriented Problems, Gesellschaft fur Mathematik und Datenverarbeiung, mbH, Institute F1/T, Postfach 1240, 5205 St. Augustin, Germany, Oct. 1991. Department of Computer Sciences & Center for Relativity ffl University of Texas at Austin Austin, TX 78712-1081

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