Method of user level parallel communication between computers based on intelligent network card

Abstract

The invention is a method for user class parallel communication among computers, which uses several high performance network which support the intelligent network card to carry on the mutual connection among commuters, adds in the device supporting the parallel communication in the system core space device driving procedure of the user class communication protocol and the user space communication bank, the data is divided and connected among the connected networks in the transmitting process. the method can realize the parallel, non-copy, sequence maintaining information transmitting among computers, thus the user can acquire the aggregation performance provided by the several connected network hardware. The method makes the parallel communication process transparent to the upper user and the bottom communication network, avoids from adding the user burden and the modification and dependents to the special network. The method is applicable to the system which uses single network can't satisfy the communication performance demands among computers.

Description

technical field [0001] The invention relates to the technical field of parallel communication between computers, in particular to a method for user-level parallel communication between computers based on an intelligent network card. Background technique [0002] In recent years, although the growth rate of computer network bandwidth exceeds the processing capacity of processors, in many cases the communication performance of computer network is still the bottleneck restricting system performance, and the communication performance of a single network cannot meet the demand. At present, most of the super servers used in the world adopt a cluster structure, and present a large-scale symmetric multiprocessor (Symmetric Multiprocessor, SMP) or cache-related non-uniform memory access (Cache Coherent Non-uniform Memory Access, CC-NUMA). Even the trend of building fleets. In this type of system, a single set of high-performance Internet is difficult to meet the communication perfor...

AI Technical Summary

Problems solved by technology

[0004] Due to the overhead of a large amount of software processing, the traditional core-level communication protocol makes its users unable to obtain the communication performance that the underlying high-performance network hardware can provide.

Although the user-level communication protocol enables its users to obtain performance close to that provided by the underlying network hardware by reducing software overhead, it is ultimately limited by the performance of the computer network hardware.

Images