A memory sharing method, heterogeneous hardware, host and heterogeneous system

Abstract

Disclosed is a memory sharing method, which is applied to heterogeneous hardware. The heterogeneous hardware is connected to a host. The method of the present invention comprises: acquiring, by heterogeneous hardware, a memory address of the partial memory of a host, which is sent by the host; and when the idle capacity of the memory of the heterogeneous hardware is lower than a predetermined capacity, according to the memory address, storing, by the heterogeneous hardware, the data in the memory of the heterogeneous hardware, which has a usage frequency lower than a predetermined usage frequency, in the partial memory. By means of the memory sharing method provided in the present invention, the data in the memory of the heterogeneous hardware, which has a usage frequency lower than a predetermined usage frequency, can be stored in the partial memory of the host, thereby reducing the storage space of the heterogeneous hardware, which is occupied by the data which has a usage frequency lower than a predetermined usage frequency, in the memory, so that the idle storage space of the heterogeneous hardware is increased, thereby providing more memory space for an application running on the heterogeneous hardware.

Description

technical field [0001] The invention relates to the field of data storage, in particular to a memory sharing method, heterogeneous hardware, a host and a heterogeneous system. Background technique [0002] An existing single-node heterogeneous system includes a host and heterogeneous hardware, and the host and hardware are interconnected through a bus PCI-E. Among them, the heterogeneous hardware may include many-core (Many Integrated Cores, MIC) expansion cards, general-purpose graphics processors (Generate-purpose Graphic Processing Units, GPGPU), etc., which serve as expansion cards to provide more powerful vector processing for the host (Host) , floating-point operations or multi-threaded parallel capabilities. Typically, in a heterogeneous system, the memory on the host is large and the memory on the heterogeneous hardware is small. Since the memory on the current heterogeneous hardware has less free storage space after storing data, it may be difficult to support the...

AI Technical Summary

Problems solved by technology

Since the current memory on heterogeneous hardware has less free storage space after storing data, it may be difficult to support the running of some applications with high memory requirements, or even to fully load the data of these applications

Images