Minyard, T., Kallinderis, Y., and Schulz, K.: Parallel load balancing for dynamic execution environments. In Proc. 34th Aerospace Sciences Meeting and  Home/Search   Document Details and Download   Summary   Related Articles   Check

....coarsening are independent of the octree, elements are not necessarily aligned with octant boundaries. Thus, choosing partition boundaries based on octants yields partitions with bumpy surfaces, which increase communication costs. This effect may be reduced by smoothing the partition boundaries [20, 29]. To do this, each processor traverses its boundary looking for elements that satisfy the following criteria: i) Four faces adjacent to four other processors. This is an isolated element that is migrated to any processor sharing a face. The donating processor s boundary is reduced by four faces, ....

....than before the exchange. The coloring may be done using subphases where a processor first sends elements to higher numbered processors and then sends them to lowernumbered ones. When three processors are involved, three subphases are necessary based on their relative order. Minyard et al. [29] perform processor boundary smoothing by a similar iterative method. They identify elements on interprocessor boundaries whose vertices are all shared by two processors. These correspond to cases (ii) iii) and (v) Patterns involving more than two processors (case (i) are not considered. After ....

T. Minyard, Y. Kallinderis, and K. Schulz. Parallel load balancing for dynamic execu- tion environments. In Proc. $Jth Aerospace Sciences Meeting and Exhibit, Reno, no. 96-0295, 1996.

.... Most of these codes are based on two dimensional finite elements [1,2,4,7,15,22] and some progress has been made towards three dimensional unstructuredmesh simulations [18,24,25] Various dynamic load balancing methods for unstructured grid applications have also been reported to date [8 10,14,19,29 31]; however, most of them lack a global view of loads across all processors. Adapt Y Repartitioning Reassignment N Y Expensive Balanced N Solution Execution N Y Finalization Remapping Initialization Partitioning Mapping Fig. 1. Overview of our global dynamic load balancing ....

T. Minyard, Y. Kallinderis, and K. Schulz, Parallel load balancing for dynamic execution environments, Proceedings 34th AIAA Aerospace Sciences Meeting (AIAA, Reno, NV, 1996) Paper 96-0295.

....This is because the computational intensity is both space and time dependent. An efficient parallel implementation of such methods is extremely difficult to achieve, primarily because of the dynamically changing nonuniform grid. Various methods on dynamic load balancing have been reported to date [5 9,11 14,16 18,24 26]; however, most of them either lack a global view of loads across processors or Proceedings of the 9th ACM Symposium on Parallel Algorithms and Architectures, Newport, Rhode Island, June 22 25, 1997. do not apply their techniques to realistic large scale applications. Figure 1 depicts our ....

T. Minyard, Y. Kallinderis, and K. Schulz, "Parallel Load Balancing for Dynamic Execution Environments," Proc. 34th AIAA Aerospace Sciences Meeting, Reno, NV, 1996, Paper 96-0295.

....leading to idle processors and adversely affecting the total execution time. An efficient parallel implementation of such methods is extremely difficult to achieve, primarily because of the dynamicallychanging nonuniform grid. Various methods on dynamic load balancing have been reported to date [7,8,10 12,14 17,19,20,22,33 35]; however, most of them either lack a global view of loads across processors or do not apply their techniques to realistic large scale applications. Balanced Repartitioning Remapping LOAD BALANCER N Expensive N Y Y MESH ADAPTOR Coarsening Edge marking Refinement FLOW SOLVER ....

Minyard, T., Kallinderis, Y., and Schulz, K. Parallel load balancing for dynamic execution environments. 34th AIAA Aerospace Sciences Meeting. AIAA, Reno, NV, 1996, Paper 96-0295.

....While these methods provide inexpensive ways to achieve a balanced computation, they can lead to degradation of partition quality. We also use an octree based partitioning, which takes advantage of an underlying tree structure to achieve balance and to maintain reasonable communication costs [17,28]. Presently, mesh refinement and coarsening precede a balancing step. Were we able to predict imbalance prior to refinement, we could maintain better performance through the enrichment and subsequent computational steps. A strategy for doing this is described in Section 3.4. Partitions often ....

....have jagged boundaries with elements penetrating into or protruding from neighboring partitions. Such features increase communication costs. As described in Section 3. 5, interprocessor boundary smoothing may be used as a post processing step to improve the quality of any load balancing procedure [19,21,28]. We solve compressible steady and transient flow problems on an IBM SP2 computer to demonstrate the capabilities of the parallel adaptive system. A steady conical flow (Example 1, Section 4.2) is used to compare load balancing procedures. The analysis of a transient shock impacting a cone ....

[Article contains additional citation context not shown here]

T. Minyard, Y. Kallinderis, and K. Schulz, Parallel load balancing for dynamic execution environments, in: Proc. 34th Aerospace Sciences Meeting and Exhibit, Reno, AIAA-96-0295 (1996).

....view of the data structures. Two basic paradigms have been used as the underlying structure for these computations. An array based approach, such as the Scalable Distributed Dynamic Array [11] uses the distribution of arrays as the fundamental unit of parallelism. The mesh based approach [1,4,14,21,24 27] uses mesh connectivity and distribution of mesh entities to achieve parallelism. Our recent efforts have focused on implementing a framework for the reliable automated solution of problems in science and engineering over arbitrary domains using scalable parallel adaptive techniques [2] This ....

....11,382 regions, and the overall MLSI=5.18 and GSI=1.69. However, since the extra effort was made to minimize communication across the slow network, this will be a very effective partitioning for this environment. A common operation when optimizing partitions is an interprocess boundary smoothing [10,12,14,21] which exchanges elements across interprocess boundaries to decrease communication volume. This operation can provide significant improvement in surface index values, especially when applied to partitions using spatial cuts to introduce a partition boundary (PSIRB or an Octreebased partitioner ....

T. Minyard, Y. Kallinderis, and K. Schulz. Parallel load balancing for dynamic execution environments. In Proc. 34th Aerospace Sciences Meeting and Exhibit, number 96-0295, Reno, 1996.

....coarsening are independent of the octree, elements are not necessarily aligned with octant boundaries. Thus, choosing partition boundaries based on octants yields partitions with bumpy surfaces, which increase communication costs. This effect may be reduced by smoothing the partition boundaries [20, 29]. To do this, each processor traverses its boundary looking for elements that satisfy the following criteria: i) Four faces adjacent to four other processors. This is an isolated element that is migrated to any processor sharing a face. The donating processor s boundary is reduced by four faces, ....

....than before the exchange. The coloring may be done using subphases where a processor first sends elements to higher numbered processors and then sends them to lowernumbered ones. When three processors are involved, three subphases are necessary based on their relative order. Minyard et al. [29] perform processor boundary smoothing by a similar iterative method. They identify elements on interprocessor boundaries whose vertices are all shared by two processors. These correspond to cases (ii) iii) and (v) Patterns involving more than two processors (case (i) are not considered. After ....

T. Minyard, Y. Kallinderis, and K. Schulz. Parallel load balancing for dynamic execution environments. In Proc. 34th Aerospace Sciences Meeting and Exhibit, Reno, no. 96-0295, 1996.

....of the octree based load balancer is examined by comparing execution times for the parallel solver before and after load balancing and by measuring the qualities of the load balanced partitions. The octree based partitioner and dynamic load balancer used for this work are discussed in [12]. The current work also uses the parallel solver and adapter presented in [13] A more thorough discussion of the finite volume solver and adapter on which the parallel implementation is based is given in [14] The generation of the hybrid grids used by this work is described in [15] ....

....of cells that are disconnected from the main region of the subdomain. The use of the octree alleviates these problems due to the structured nature of the octree and the biasing of the octree to the geometry of interest. A more detailed explanation of the octree based partitioner is given in [12] The structure of the prisms in the normal tothe surface direction is exploited. The prisms are defined by their corresponding base faces on the surface. As a result, the stacks of prisms are partitioned by simply partitioning the triangular surface mesh. All cells within each prism stack are ....

[Article contains additional citation context not shown here]

T. Minyard, Y. Kallinderis, and K. Schulz, "Parallel Load Balancing for Dynamic Execution Environments," AIAA Paper 96-0295, Reno, NV, January 1996.

No context found.

Minyard, T., Kallinderis, Y., and Schulz, K.: Parallel load balancing for dynamic execution environments. In Proc. 34th Aerospace Sciences Meeting and

No context found.

Minyard, T. and Kallinderis, Y.: Parallel load balancing for dynamic execution environments. Comput. Methods Appl. Mech. Engrg., 189(4):1295--1309, (2000)

No context found.

T. Minyard, Y. Kallinderis, Parallel load balancing for dynamic execution environments, Comput. Methods Appl. Mech. Engrg. 189 (4) (2000) 1295--1309.

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