ABOUT PARALLELIZATION OF SOLVING OF BOUNDARY VALUE PROBLEMS ON QUASISTRUCTURED GRIDS

Abstract

Technological components of boundary problems solution on offered quasi-structured grids of special  kind  is  considered.  The  feature  of  these  grids  is  that  both  macrogrid  (coarse  grid)  in  a whole domain and subgrids (local grids) in subdomains are structured and rectangular, it provides efficient  structure  of  data  and  effective  using  of  computational  algorithms.  At  the  same  time, resulting  quasi-structured  grid  is  adaptive  to  irregularities  within  a  domain  and  to  complicated shape of domain boundary. It is essential that subgrids can be unmatched. One variant of domain
decomposition  methods  for  solving  boundary  problems is  offered,  this  one  is  based  on  separate approximation  of  boundary  problem  on  the  interface  and  within  the  subdomains.  In  order  to balance  utilization  of  processors  whole  set  of  subdomains  is  divided  into  unions  (groups)  of subdomains. Estimates of paralellization efficiencywas obtained for model problem using different number of processors, different grids and differentunions of subdomains.