Load balancing of processors when solving the problems of fluid and gas mechanics by mesh methods

Numerical solution of problems of fluid and gas mechanics on multiprocessor computing systems involves a geometric decomposition of the computational domain, handling the corresponding subdomain by each processor, and communications between processors for a complete solution. Load balancing of processors is specified by the uniformity of the mesh distribution between processors and the cost of data transfer between processors. The cost of data transfer between processors depends on the number of connections between the subdomains distributed over the processors. Approaches to the static and dynamic load balancing of processors are considered to solve the problems of fluid and gas mechanics on multiprocessor computing systems. Various stages and methods of static (methods of bisection, combinatorial methods, combined approaches) and dynamic (diffusive algorithm, method of potential, multilevel approaches) load balancing are discussed, and their performance indices are compared. The diffusive method and the method of potential are compared for a domain of simple geometric configuration to solve the problem on an adaptive grid.