Autonomic Dynamic Load Balancing of Parallel SAMR Applications
Independent thesis Advanced level (professional degree), 20 credits / 30 HE creditsStudent thesis
When solving partial differential equations using finite difference methods on structured meshes, adaptive refinement can be used to increase the accuracy of the solution in an efficient manner. When implementing solvers using structured adaptive mesh refinement for modern parallel computer systems, an important task is the partitioning of the grid hierarchy over the available processors. The Meta-partitioner is an autonomic framework which can dynamically select between a large number of grid-partitioning algorithms at run time.
In this thesis we investigate which modifications that are necessary in order to connect the Meta-partitioner to the existing SAMR-framework Chombo, and begin the process of performing this connection. We conclude that although significant changes to both Chombo and the Meta-partitioner are necessary, a connection definitely seams feasible. We estimate that that the major work of the connection has been done, and that with the experience gained from this project, the continuation is straightforward.
We also connect a patch-based partitioning algorithm to Chombo and evaluate it for the first time as part of a real SAMR-based simulation. The results are promising and we conclude that it is a viable candidate for inclusion in the Meta-partitioner.
Place, publisher, year, edition, pages
2011. , 35 p.
UPTEC F, ISSN 1401-5757 ; 11041
SAMR, parallel, adaptive, refinement, structured mesh, partitioning, Chombo, Meta-partitioner
IdentifiersURN: urn:nbn:se:uu:diva-155517OAI: oai:DiVA.org:uu-155517DiVA: diva2:426396
Master Programme in Engineering Physics
Thuné, MichaelRantakokko, Jarmo
Holmgren, SverkerNyberg, Tomas