Traffic distribution method, device, computer equipment and storage medium

Abstract

The embodiment of the present invention discloses a traffic distribution method, device, computer equipment and storage medium, comprising the following steps: acquiring a receiving end adjustment value of a network traffic data packet to be processed; performing a hash operation on the receiving end adjustment value obtaining the hash value of the network traffic data packet; searching for a target queue corresponding to the hash value in a plurality of preset queues according to the hash value and adding the network traffic data packet to the target queue , wherein the value of the hash value is within the value range of the number of the plurality of queues. The hash value of the network data packet is obtained by hashing the mediation value of the receiving end, and the target queue is searched in the preset multiple queues by using the hash value, which can ensure that the network traffic data packets are evenly distributed among the multiple queues. , to avoid the problem of system processing performance degradation caused by excessive load or idle of processor cores.

Description

technical field [0001] The embodiments of the present invention relate to the technical field of computers, and in particular, to a flow distribution method, device, computer equipment, and storage medium. Background technique [0002] RSS (receive side scaling) is a load distribution method proposed by Microsoft. By calculating the network layer and transport layer 2 / 3 / 4 tuple HASH value in the network data message, the least significant bit (LSB) of the HASH value is taken. It is used to index the indirect addressing table RETA (Redirection Table), and the saved index value in the indirect addressing table RETA is used to allocate data packets to different CPU COREs for receiving and processing. At this stage, RSS has basically been implemented in hardware. Through this technology, network traffic can be offloaded to multiple CPU COREs, reducing the occupancy rate of a single CPU in the operating system. [0003] In order to achieve high-performance processing of network ...

AI Technical Summary

Problems solved by technology

[0004] However, the inventor has found that each hardware queue of each network card corresponds to a processing CORE. When the number of network card hardware queues is small and the number of CPU COREs is large, there are too many idle cpu resources; NIC_B with a small number of queues. At this time, the number of COREs occupied by traffic processing is based on NIC_A. At this time, the traffic of NIC_A and NIC_B will be hashed to the same CORE, which will increase the load of these COREs, and some COREs only process the traffic of NIC_A. , traffic cannot be evenly hashed to different CPU COREs, resulting in failure to improve overall system performance

Moreover, the experiment also found that when the number of network card hardware queues is configured too much, the system performance will decrease instead.

Therefore, the above situation will lead to the need to solve the problem that the number of network card hardware queues does not match the number of parallel processing COREs, and the traffic cannot be evenly hashed to multiple COREs, resulting in system performance degradation.

Images