Parallelization of the Multigrid Solver With Processor Virtualization
Processor virtualization is a parallelization technique proposed to get rid of the limitation of physical processors, while exploring the natural parallelism of the application. The virtualization of the processors enables the system to do run-time load balancing automatically, which is critical for some complex and dynamic applications, for example, the molecular dynamic simulation. This paper investigates the parallelization of the multigrid solver using the processor virtualization technique. This paper uses a neighbor based virtual processors to physical processors mapping strategy and dynamically changing the number of virtual processors when reaching the critical level.