J. Clinckemaillie, B. Elsner, G. Lonsdale, S. Meliciani, S. Vlachoutsis, F. de Bruyne, and M. Holzner. Performance Issues of the Parallel PAM-CRASH Code. Int. J. Supercomputer Appl. & High Performance Comput., 11(1):3--11, 1997.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....broad class of graph manipulation approaches. 3.6.2. Contactimpact simulations One of the particular areas of interest driving the development of multiphase partitioning algorithms has been the use of contact impact algorithms (for example in the automotive industry for simulating crashes, e.g. [5]) Typically the simulation will involve localised stress strain finite element calculations over the entire mesh together with a much more complex contact impact detection phase over the restricted areas of possible penetration, 20] It is usually the imbalance introduced during this contact ....

....introduced during this contact phase which is responsible for serious deterioration in the overall scalability of the code and several approaches to overcome it have been tried. Vertex veighting. One strategy arising from crashworthiness simulations and designed by Clinckemaillie, Lonsdale et al. [5,20], to address this problem, is to use a static partition of the mesh but to add contact related weights to the partitioning cost function for vertices that are part of a contact surface. Although this approach does not directly address the two phase nature of the problem, a significant improvement ....

[Article contains additional citation context not shown here]

J. Clinckemaillie, B. Elsner, G. Lonsdale, S. Meliciani, S. Vlachoutsis, F. de Bruyne, M. Holzner, Performance issues of the parallel PAM-CRASH code, Int. J. Supercomputer Appl. High Performance Comput. 11 (1) (1997) 311.

....element and per node computational costs and element element, node node, and element node data dependencies for communication. In addition to data dependencies between neighbouring elements and nodes in the mesh, dependencies between arbitrary parts of the mesh can occur. For the PAM CRASH code [5], such data dependencies originate within the contact impact algorithms when the penetration of mesh segments by non connected nodes is detected and corrected. The DRAMA cost model allows the construction of virtual elements [2, 9, 10] which represent the occurring costs of such dependencies. ....

....communication costs easier to specify and to handle than element based communication costs. In fact, the exchange of node information is usually the method of choice for sub domain interface communications in parallel FE application codes; examples are the current versions of PAM CRASH and FORGE3 [5, 6]. If communication costs arise between elements and nodes then a third parameter, noc (u) is required to determine the cost function completely. We suppose that node information is sent to the corresponding elements, and after a calculation step involving information from more than one node ....

J. Clinckemaillie, B. Elsner, G. Lonsdale, S. Meliciani, S. Vlachoutsis, F. de Bruyne, and M. Holzner, Performance Issues of the Parallel PAMCRASH Code, The International Journal of Supercomputer Applications and High Performance Computing, Vol. 11, No. 1, pp. 3-11, 1997.

.... LaRIA Universit de Picardie Jules Vernes 80000 Amiens, France paladin laria.u picardie.fr November 18, 1998 1 Introduction Many of important computational applications involve solving problems with very large data sets [16] For example astronomical simulation [17] crash test simulation [3], global climate modeling [6] and many other scienti c and engineering problems can involve data sets that are too large to t in main memory. Using parallelism can reduce the computation time and increase the available memory size, but for challenging applications the memory is ever insucient in ....

J. Clinckemaillie, B. Elsner, G. Lonsdale, S. Meliciani, S. Vlachoutsis, F. de Bruyne, and M. Holzner. Performance issues of the parallel PAM-CRASH code. The International Journal of Supercomputer Applications and High Performance Computing, 11(1):3-11, Spring 1997.

....and one library (MMUM) and run on Linux. We present a new pagination strategy for the LU decomposition program. 1 Introduction Many of important computational applications involve solving problems with very large data sets [14] For example astronomical simulation [15] crash test simulation [3], global climate modeling [6] and many other scienti c and engineering problems can involve data sets that are too large to t in main memory. Using parallelism can reduce the computation time and increase the available memory size, but for challenging applications the memory is ever insucient in ....

J. Clinckemaillie, B. Elsner, G. Lonsdale, S. Meliciani, S. Vlachoutsis, F. de Bruyne, and M. Holzner. Performance issues of the parallel PAM-CRASH code. The International Journal of Supercomputer Applications and High Performance Computing, 11(1):3-11, Spring 1997.

.... IO bandwidth and the computation rate: There is no need of large main memory for the factorization of huge matrix 1 Introduction Many of important computational applications involve solving problems with very large data sets [7] For example astronomical simulation [8] crash test simulation [3], global climate modelling, and many other scienti c and engineering problems This work is supported by a grant of the P ole de Mod elisation de la R egion Picardie . can involve data sets that are too large to t in main memory. Using parallelism can reduce the computation time and increase ....

J. Clinckemaillie, B. Elsner, G. Lonsdale, S. Meliciani, S. Vlachoutsis, F. de Bruyne, and M. Holzner. Performance issues of the parallel PAM-CRASH code. The International Journal of Supercomputer Applications and High Performance Computing, 11(1):3-11, Spring 1997.

....own a node of this element. As a consequence, redundant computations are introduced, with an overhead depending on the computational costs of the function Work. 5 Performance Results For the evaluation of the exclusive ownership technique we used a kernel from an industrial crash simulation code [5]. The kernel is based on a time marching scheme to perform stress strain calculations on a finite element mesh consisting of 4 node shell elements. In each time step elemental forces are calculated for every element of the mesh (cf. function Work) and added back to the forces stored at nodes by ....

J. Clinckemaillie, B. Elsner, and G. Lonsdale et al. Performance issues of the parallel PAM-CRASH code. The International Journal of Supercomputer Applications and High Performance Computing, 11(1):3--11, Spring 1997.

....s 1.021 s 0.545 s HPF SM 1.940 s 0.983 s 0.500 s 0.264 s Table 2. Unstructured relaxation on NEC SX 4 (1M grid points, 73 iterations) Crash Simulation Kernel For the evaluation of a scientific computation with unstructured reductions, we used a kernel from an industrial crash simulation code [7]. The kernel is based on a time marching scheme to perform stress strain calculations on a finite element mesh consisting of 4 node shell elements. In each time step elemental forces are calculated for every element of the mesh and added back to the forces stored at nodes by means of unstructured ....

J. Clinckemaillie, B. Elsner, and G. L. et al. Performance issues of the parallel PAM-CRASH code. The International Journal of Supercomputer Applications and High Performance Computing, 11(1):3--11, Spring 1997.

....and socio economic systems. Examples of such simulations include weather modeling and forecasting, behavior of subatomic particles, crash tests in automobile industry, optimizing combustion engines, aerodynamic behavior of airplanes, study of chemical phenomena, or financial modeling (e.g. see [8, 14, 37, 75, 64]) Solving these problems requires appropriate hardware and software support. 1.1.1 Hardware Architectures of HPCs From the architectural point of view HPCs offer computational concurrency and may take different forms. Above all, parallel computers may differ in control mechanism, memory ....

J. Clinckemaillie, B. Elsner, G. Lonsdale, S. Meliciani, S. Vlachoutsis, F. de Bruyne, and M. Holzner. Performance issues of the parallel PAM-CRASH code. Journal of Supercomputing Applications and High-Performance Computing, 11(1), 1997.

....Although HPF 1 was very well suited for applications based on regular data structures and access patterns, it soon became apparent that it did not provide enough flexibility for an efficient formulation of many advanced algorithms. For example, weather forecasting codes[2] or crash simulations[19], are often characterized by the need to distribute data in an irregular manner and dynamically balance the computational load of the processors. The current version of HPF, HPF 2[46] and in particular its approved extensions , is a significant step in supporting such algorithms. The study of ....

J. Clinckemaillie, B. Elsner, G. Lonsdale, S. Meliciani, S. Vlachoutsis, F. de Bruyne, and M. Holzner. Performance issues of the parallel PAM-CRASH code. The International Journal of Supercomputer Applications and High Performance Computing, 11(1):3--11, Spring 1997.

....broad class of graph manipulation approaches. 3.6. 2 Contact impact simulations One of the particular areas of interest driving the development of multiphase partitioning algorithms has been the use of contact impact algorithms (for example in the automotive industry for simulating crashes, e.g. [5]) Typically the simulation will involve localised stress strain finite element calculations over the entire mesh together with a much more complex contact impact detection phase over the localised areas of possible penetration, 20] It is usually the imbalance introduced during this contact ....

....during this contact phase which is responsible for serious deterioration in the overall scalability of the code and several approaches to overcome it have been tried. Vertex weighting. One strategy arising from crashworthiness simulations and designed by Clinckemaillie, Lonsdale et al. [5, 20], to address this problem, is to use a static partition of the mesh but to add contact related weights to the partitioning cost function for vertices that are part of a contact surface. Although this approach does not directly address the two phase nature of the problem, a significant improvement ....

[Article contains additional citation context not shown here]

J. Clinckemaillie, B. Elsner, G. Lonsdale, S. Meliciani, S. Vlachoutsis, F. de Bruyne, and M. Holzner. Performance Issues of the Parallel PAM-CRASH Code. Int. J. Supercomputer Appl. & High Performance Comput., 11(1):3--11, 1997.

....element and per node computational costs and element element, node node, and element node data dependencies for communication. In addition to data dependencies between neighbouring elements and nodes in the mesh, dependencies between arbitrary parts of the mesh can occur. For the PAM CRASH code [5], such data dependencies originate within the contact impact algorithms when the penetration of mesh segments by non connected nodes is detected and corrected. The DRAMA cost model allows the construction of virtual elements [2, 9, 10] which represent the occurring costs of such dependencies. ....

....communication costs easier to specify and to handle than element based communication costs. In fact, the exchange of node information is usually the method of choice for sub domain interface communications in parallel FE application codes; examples are the current versions of PAM CRASH and FORGE3 [5, 6]. If communication costs arise between elements and nodes then a third parameter, noc en (u) is required to determine the cost function completely. We suppose that node information is sent to the corresponding elements, and after a calculation step involving information from more than one node ....

J. Clinckemaillie, B. Elsner, G. Lonsdale, S. Meliciani, S. Vlachoutsis, F. de Bruyne, and M. Holzner, Performance Issues of the Parallel PAMCRASH Code, The International Journal of Supercomputer Applications and High Performance Computing, Vol. 11, No. 1, pp. 3-11, 1997.

....passing programs have been compiled with the SUN F90 1.2 compiler. In Figures 1 and 2 the performance of these kernels is compared to hand written MPI message passing programs. The first kernel, developed by Guy Lonsdale, represents the basic stressstrain calculation of a crash simulation code [13] based on 4 node shell elements. The kernel used in our evaluation employed an unstructured mesh with 25760 nodes and 25600 elements. The total size of the kernel is approximately 600 lines of code. It uses an explicit time marching scheme which is represented by an outer time step loop performing ....

J. Clinckemaillie, B. Elsner, G. Lonsdale, S. Meliciani, S. Vlachoutsis, F. de Bruyne, M. Holzner. Performance Issues of the Parallel PAM-CRASH Code. Journal of Supercomputing Applications and High-Performance Computing, Vol.11, No.1, 1997.

....The DRAMA library aims at the provision of tools to achieve dynamic load balancing for unstructured grid oriented problems on distributed memory systems that support message passing. In this report, the interface between DRAMA library modules and the application codes PAM CRASH STAMP [4] and FORGE3 [5] considered in the DRAMA project is described. The basis is the preliminary DRAMA cost model from [3] 2 DRAMA Library Modules The basic modules that the DRAMA library will contain include: 1. COST FUNCTION EVALUATION based on the cost monitor parameters 2. GRAPH CONSTRUCTION 3. ....

....quality. It is the only possibility if the application program results in communications between elements and nodes and extreme accuracy in the cost function and quality of the partition is required. Take for example the stress strain calculations without contact impact computations in PAM CRASH [4]. The computations are performed in four steps (see figure 9) ffl Element strains calculated from nodal coordinates. ffl Element stresses calculated from element strains. ffl Nodal forces calculated from element stresses. F 1 ; F 2 ; F 3 ; F 4 F i = F i F i ffl Nodal ....

[Article contains additional citation context not shown here]

J. Clinckemaillie, B. Elsner, G. Lonsdale, S. Meliciani, S. Vlachoutsis, F. de Bruyne, and M. Holzner, Performance Issues of the Parallel PAMCRASH Code, The International Journal of Supercomputer Applications and High Performance Computing, Vol. 11, No. 1, pp. 3-11, 1997.

.... VARIABLE section: declare global variables used by DRAMA int DRAMACOMM; int DRAMANecalc[DRAMANeumax] int DRAMANecom[DRAMANpmax] DRAMANevmax] DRAMANeumax] int DRAMANncalc[DRAMANnumax] int DRAMANncom[DRAMANpmax] DRAMANnumax] int DRAMANencom[DRAMANpmax] DRAMANnumax][2]; float DRAMAnope[DRAMANeumax] float DRAMAnoce[DRAMANevmax] DRAMANeumax] float DRAMAnopn[DRAMANnumax] float DRAMAnocn[DRAMANnumax] float DRAMAnocen[DRAMANnumax] double DRAMAdt[ DRAMAdtdim ] float DRAMAthreshold; float DRAMALlb[DRAMANpmax] DRAMAMAXSTEP] 3] int DRAMAstatus[ DRAMAstatusdim ] ....

J. Clinckemaillie, B. Elsner, G. Lonsdale, S. Meliciani, S. Vlachoutsis, F. de Bruyne, and M. Holzner, Performance Issues of the Parallel PAMCRASH Code, The International Journal of Supercomputer Applications and High Performance Computing, Vol. 11, No. 1, pp. 3-11, 1997.

....a fully parallel mesh generator has been developed that exploits the parallel re partitioning of adaptively generated meshes. The mesh re allocation approach to dynamic load balancing has been demonstrated and validated by the leading industrial codes PAM CRASH (for crashworthiness simulation, [5,6]) PAM STAMP (for metal stamping deep drawing and related simulations [7,8] FORGE3 (for forging with viscoplastic incompressible materials, 9,10] Section 3 will present parallel performance behaviour for the above applications using the DRAMA library. Despite this emphasis on the validation ....

....around 90 of stamping applications rely on the adaptive meshing features. In contrast to the re meshing approach adopted by FORGE3, PAM STAMP uses a mesh refinement (and coarsening) strategy based on the original user defined mesh. Leaving details of the algorithms and their parallelisation to [5,6] and the references therein, a summary of the solution methods of the PAM SOLID based codes is as follows: an explicit timeintegration; a non linear finite element method using a Lagrangian unstructured mesh; elementwise stress strain calculations supplemented by penalty method contact impact ....

J. Clinckemaillie, et al., Performance issues of the parallel PAM-CRASH code, Int. J. Supercomput. Appl. High Performance Comput. 11 (1) (1997) 3 11.

....the necessity for multi phase multi constraint re partitioning components [9,10,16] for cost re balancing. These new partitioning methods allow load balancing for several computational phases e.g. stress strain analysis and contact treatment in the industrial crash analysis code PAM CRASH TM [6] that are separated by synchronisation points. In Section 2, we present the basic features of this cost model which allows a general approach to load identification, modelling and imbalance minimisation. Section 3 briefly describes the library interface, Section 4 treats the principles of ....

....a list of nodes per element which is a native data structure (element connectivity) in most finite element appli cations. In addition to data dependencies between neighbouring elements and nodes in the mesh, dependencies between arbitrary parts of the mesh can occur. For the PAM CRASH TM code [6], such data dependencies originate within the contact impact algorithms when the penetration of mesh segments by non connected nodes is de tected and corrected. The DRAMA cost model allows the construction of virtual elements [4,12,13] which represent the occurring costs of such dependencies (see ....

J. Clinckemaillie, B. Elsner, G. Lonsdale, S. Meliciani, S. Vlachoutsis, F. de Bruyne, M. Holzner, Performance issues of the parallel PAM-CRASH code, Int. J. Supercomputer Applications and High Performance Computing 11 (1997) 3 11.

....induce only small changes in the partitions. Most partitioning algorithms do not exhibit the latter behaviour. Third, the costs for the RCB algorithm itself are low and finally, it is easy to parallelize [36, 37] 5. 2 Two Mesh Approach with RCB Transient solid dynamics codes such as PAM CRASH [33] or PRONTO [31, 37] consist of two computational phases: finite element (FE) analysis and contact detection correction. The practical importance of these calculations (car crashes, metal forming, cutting for manufacturing processes) and their computational intensiveness make them natural ....

....compute nodes of the Sandia Intel Teraflop computer. In contrast to PRONTO, the FE mesh of PAM CRASH is a surface mesh mainly consisting of shell elements. Thus the PRONTO two mesh approach is not applicable. Contact areas, however, can be specified by users; these areas are called slide lines [33]. Since for a frontal crash, e.g. contact most probably occurs only in the frontal part of a car a suitably chosen slide line in the frontal part could serve as second mesh, the contact mesh. This partial FE mesh could then be partitioned dynamically depending on the contact behaviour in the ....

J. Clinckemaillie, B. Elsner, G. Lonsdale, S. Meliciani, S. Vlachoutsis, F. de Bruyne, and M. Holzner, Performance Issues of the Parallel PAM-CRASH Code, The International Journal of Supercomputer Applications and High Performance Computing, Vol. 11, No. 1, 1997, pp. 3-11.

....mesh refinement (and coarsening) strategy based on the original user defined mesh. The current message passing versions have been further developed from the prototypes produced within the EUROPORT ( 24] and EUROPORT D ( 15] projects. Leaving details of the algorithms and their parallelisation to [6,19] and the references therein, a summary of the solution methods of the PAM SOLID based codes is as follows: an explicit timeintegration; a non linear finite element method using a Lagrangian unstructured mesh; elementwise stress strain calculations supplemented by penalty method contact impact ....

J Clinckemaillie, B Elsner, G Lonsdale, S Meliciani, S Vlachoutsis, F de Bruyne and M Holzner, Performance issues of the parallel PAM-CRASH code, Int. J. Supercomputer Applications and High Performance Computing, 11(1), 3-11,1997

....adaptive meshing features. In contrast to the re meshing approach adopted by FORGE3, PAMSTAMP uses a mesh refinement (and coarsening) strategy based on the original user defined mesh. An example of such a mesh can be seen in Figure 2. Leaving details of the algorithms and their parallelisation to [6,7] and the references therein, a summary of the PAM SOLID based codes is as follows: The non linear explicit finite element method employed uses a Lagrangian formulation of the equations of motion of the nodes of the unstructured mesh constructed by the replacement of the physical model by an ....

J Clinckemaillie, B Elsner, G Lonsdale, S Meliciani, S Vlachoutsis, F de Bruyne and M Holzner, Performance issues of the parallel PAM-CRASH code, Int. J. Supercomputer Applications and High Performance Computing, 11(1), 3-11,1997

No context found.

J. Clinckemaillie, B. Elsner, G. Lonsdale, S. Meliciani, S. Vlachoutsis, F. de Bruyne, and M. Holzner. Performance Issues of the Parallel PAM-CRASH Code. Int. J. Supercomputer Appl. & High Performance Comput., 11(1):3--11, 1997.

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