An Efficient Long Message Oriented Computer Network Communication Method

Abstract

An efficient long-message-oriented computer network communication method. When processes between different computers need to transmit long messages, the following steps are performed: ①Reserve a small part of physical memory as a shared communication buffer; ②The sender process starts from the The reserved free physical memory applies for the shared communication buffer; 3. Map the network interface to the user space, so that the sender process can directly access the network interface; the sender process copies the data to be sent to the shared communication buffer; After the data to be sent is copied to the shared communication buffer, the sender's network card writes the data in the sender's shared communication buffer into the network card's sending buffer through DMA; the receiver's network card requests the communication software after part of the data is written into the shared communication buffer. Processing; the communication software sends a signal that the data has been received to the upper-level receiver process, requesting the receiver process to process; the receiver process copies the data in the shared communication buffer to the destination address. The technical effects of the present invention are as follows: the overhead of the network card starting DMA multiple times is reduced, and the communication rate of long messages is significantly accelerated; and the parallel method of user mode memory copy and network card DMA pipeline is adopted to reduce the cost of memory copy.

Description

technical field [0001] The invention relates to a communication method, in particular to an efficient long message-oriented computer network communication method. Background technique [0002] In order to solve the problem of copying data to the communication buffer many times in the process of high-speed network transmission of large data, most systems currently use user-mode communication mechanisms and "zero-copy" technology. That is, during data transmission, the virtual address of the user mode is converted into the corresponding physical address, and it is locked, and then the physical address is passed to the network card for DMA (Direct Memory Access, direct memory access) data transmission. [0003] The above-mentioned method realizes zero copy of data in the communication process, and eliminates the copying time of data in memory. However, in the case of high system load (for example, CPU and memory utilization are above 90%), the communication efficiency is abou...

AI Technical Summary

Problems solved by technology

The reason is: when the system memory utilization rate is high, the large continuous physical memory space of the entire physical memory is very small, and most of them are some scattered pages. The real location of the memory is very scattered, that is, the physical memory page where the message is located is discontinuous, so when the network card uses DMA to send the message, it is necessary to start DMA multiple times to read data from the memory, which is very difficult for the sender. Time-consuming; in the same way, when the system memory utilization rate is high and the received messages are relatively large, the receiver application can only provide most of the scattered physical memory pages to the network card, so the receiver's network card must also need Start DMA multiple times to write data into memory, which is also time-consuming on the receiving side

Images