Tools to support mesh adaptation on massively parallel computers

摘要: The scalable execution of parallel adaptive analyses requires the application dynamic load balancing to repartition mesh into a set parts with balanced work and minimal communication. As meshes being generated reach billions elements are performed on massively computers 100,000’s computing cores, number complexities arise that need be addressed. This paper presents procedures developed deal two them. first is procedure support multiple per processor which used as increases in size it desirable partition larger cores than currently used. second predictive method sets entity weights before steps so well after adaptation step thus avoiding excessive memory spikes would otherwise occur during adaptation.