GPU-based network function virtualization system, method and storage medium

Abstract

The invention discloses a network function virtualization system and method based on GPU and storage medium, and the system comprises an SFC control module (1) and an SFC agent module (2) disposed in the GPU, an SFC (3) is packaged in the SFC agent module (2), and each SFC control module (1) corresponds to one SFC agent module (2); the SFC control module (1) comprises: an adaptive module (11) for aggregating a plurality of received data packets to obtain an aggregated data packet and calculating the number of computing resources allocated to a GPU of the aggregated data packet; an SFC starting module (12) is used for calling the SFC agent module (2) in the GPU; the SFC agent module (2) is used for starting an SCF (3) according to the calling of the SFC starting module (12); and the SCF (3) is used for receiving the aggregated data packet sent by the self-adaptive module (11), processing the aggregated data packet by adopting the computing resources of the GPU, obtaining the processed data packet and sending the processed data packet to the SFC control module (1).

Description

technical field [0001] The invention relates to network function virtualization technology, in particular to a GPU-based network function virtualization system and method and a computer-readable storage medium. Background technique [0002] (1) Network function virtualization technology [0003] In a traditional network, network functions are carried by proprietary hardware, and these network functions are called middleware (Middlebox). These middleware are very common in network infrastructure, such as firewalls, intrusion detection systems (Intrusion Detection System, IDS), network optimization agents, network protocol security (Internet Protocol Security, IPSec), and so on. However, traditional network functions developed based on proprietary hardware have a series of problems: First, with the development of software-defined network (Software Defined Network, SDN) architecture, network innovation becomes simple, while network functions driven by traditional proprietary h...

AI Technical Summary

Problems solved by technology

However, traditional network functions developed based on proprietary hardware have a series of problems: First, with the development of software-defined network (Software Defined Network, SDN) architecture, network innovation becomes simple, while network functions driven by traditional proprietary hardware cannot The characteristics of rapid update and iteration will reduce the speed of network innovation; secondly, as the scale of the network gradually increases, for example, the traffic of the data center will fluctuate frequently over time, which requires flexible and dynamic expansion of network functions; finally, SDN provides network management Brings a better and more flexible way, while traditional network functions do not enjoy this kind of operation and maintenance

[0007] (3) Insufficiency of existing network function elastic scaling solutions

However, since each data packet needs to access the remote state, compared with the original state storage scheme, it will bring additional processing time to each data packet, affecting end-to-end delay and throughput

[0010] The root cause of the technical problems in the existing technology is that the current NFV system based on the general-purpose CPU platform can only provide very limited computing resources, and the CPU is usually allocated in units of cores, which is not conducive to fine-grained resource management

Images