System and method for controlling communication middleware among internal multi-class computing units

Abstract

The invention belongs to the technical field of heterogeneous high-performance computing, and discloses a system and method for controlling the communication middleware among internal multi-class computing units. The computing task is matched with the computing units and the communication middleware according to the the business characteristics; and the middleware is specially used for solving theproblems that the data interaction interfaces among the computing units of various different structures in a heterogeneous system are inconsistent and the DMA cannot be configured for part of the computing units. The middleware can be used for replacing the part of computing units to configure DMA, and the DMA is uniformly used among various heterogeneous computing units to construct an intercommunication link, so that the data interaction is directly completed without the main memory of the system, a large number of memory operations are reduced, and the problem that a heterogeneous computing system is higher in computing time delay is solved, and the heterogeneous computing system which is low in time delay and dynamically and elastically adjusted is provided for a user so as to meet the high-performance computing requirements in the fields of spaceflight, medical treatment, e-commerce, artificial intelligence, entertainment and the like.

Description

technical field [0001] The invention belongs to the technical field of heterogeneous high-performance computing, and in particular relates to a communication middleware control system and method between internal multi-category computing units. Background technique [0002] At present, multiple computing components are integrated in a heterogeneous system at the same time, which combines the advantages of multiple heterogeneous computing units, and its computing power is far stronger than that of traditional computing systems. However, there are also certain problems in heterogeneous systems. The various types of computing units in the system have different architectures, and there are obstacles in data exchange between different types of computing units, resulting in high computing delays in the entire heterogeneous system. At present, there are mainly two mainstream technologies to solve the problem of high computing delay in this heterogeneous system: first, the main memor...

AI Technical Summary

Problems solved by technology

This method only supports specific heterogeneous computing units (such as FPGA), and the application area is narrow and the operation is complex

[0003] To sum up, the problem existing in the prior art is that each heterogeneous computing unit needs to occupy a large amount of system main memory and involves

Images