Design of Multiple PVAS on InfiniBand Cluster System Consisting of Many-core and Multi-core
Mikiko Sato, Go Fukazawa, Akio Shimada, Atsushi Hori, Yutaka Ishikawa, Mitaro Namiki · 2014
This paper proposes a novel parallel task model for an exascale heterogeneous computing environment consisting of multi-core and many-core CPUs. In the proposed environment, "Multiple Partitioned Virtual Address Space (Multiple PVAS)," all PVAS tasks on the hybrid system share the same virtual address space, so any PVAS task can access the data owned by the other PVAS tasks no matter on which CPU they are running. Multiple PVAS is implemented in the Linux Kernel on Xeon and Xeon Phi. The basic performance of the Multiple PVAS shows that it is effective for the communication between tasks on the hybrid system.