RDMA-based data transmission method and RDMA network cards

Abstract

An embodiment of the invention provides an RDMA-based data transmission method and RDMA network cards, and relates to the communication field, so that complicated bus conversion and multistage module transfer process in the process where the RDMA network card transmits a message to a central processing unit can be simplified, and message transmission efficiency between the RDMA network card and the central processing unit is improved. The method comprises the following steps: the RDMA network card of a first node receives an RDMA reading request sent by the RDMA network card of a second node, wherein the RDMA reading request comprises memory address of data requested for reading by the RDMA reading request in a memory of the first node; carrying out analysis on the RDMA reading request to obtain the memory address, and generating an ACE reading request; sending the ACE reading request to a central processing unit of the second node; receiving a first ACE response message returned by a central processing unit of the first node; analyzing the first ACE response message to obtain the data requested for reading by the RDMA reading request, and generating an RDMA response message; and sending the RDMA response message to the RDMA network card of the second node. The method is used for data transmission between the nodes.

Description

technical field [0001] The invention relates to the communication field, in particular to an RDMA-based data transmission method and an RDMA network card. Background technique [0002] Remote direct memory access (Remote Direct Memory Access, referred to as RDMA) is a technology for direct remote memory access, that is, data can be directly and quickly migrated from one system to another remote system memory without any impact on the operating system, reducing It eliminates the consumption of the central processing unit participating in the data transmission process, liberates the memory bandwidth and processor (Central Processing Unit, referred to as CPU) cycle, and then improves the performance of the system processing business, with the characteristics of high bandwidth, low latency and low CPU usage. [0003] Because the bus interface of the network card with RDMA function (referred to as RDMA network card) on the market generally adopts the PCIE (PCI-Express) standard i...

AI Technical Summary

Problems solved by technology

[0004] However, in the process of realizing the above-mentioned data migration, the inventor found that since the RDMA network card using the PCIE interface cannot directly access the memory, it is necessary to perform relevant protocol conversion and component delivery on the request before the CPU or memory Generate a response, that is, convert the request message supporting the RDMA protocol into a request message supporting the PCIE protocol, and then convert the request message supporting the PCIE protocol into a bus transaction message, that is, a protocol that supports direct access to the CPU or memory request message, which makes the entire data migration process too complicated and increases the delay of the system

At the same time, since the RDMA network card is not responsible for the data consistency before and after the protocol conversion, it needs to be processed by the CPU, which increases the workload of the CPU.

Images