The research is driven by SFC index computation free dynamic adaptivity, Cluster based simulations as well as Cluster based load balancing and data migration for shared- and distributed memory systems.
With our novel cluster based parallelization, our domain is split into several non-overlapping chunks of connected cells. The heuristic we employ for efficient decomposition is based on space-tree splits along the Sierpiński SFC, thus its applicability is also given for some other SFCs. The communication pattern used in this framework offer highly-efficient communication buffers by dynamically updated run-length encoded communication patterns for edges and vertices shared by adjacent cluster. Furthermore, highly efficient data migration due to quasi-optimal compact storage of cluster payload and communication data stored implicitly is implemented.
The goal is not only to give a convenient way to express this kind of problems (aka software engineering) but also to consider aspects of efficiency as well as scalability (aka HPC computing).
The features include:
https://www5.in.tum.de/sierpinski/
No comments:
Post a Comment