Processing method for real-time network streaming applications using GPU acceleration

Abstract

The invention discloses a scheduling method using GPU to accelerate real-time network stream processing, which is characterized in that in the method, the CPU sets the GPU execution cycle I according to the application program that needs to be accelerated; the GPU execution cycle I is greater than or equal to the worst time of the GPU Processing time; including the following steps: (1) The CPU receives the calculation results of the GPU in the previous GPU execution cycle I, and at the beginning of the next batch processing time cycle, the CPU starts to accumulate data or execution sequences in the GPU execution cycle I; ( 2) When the next batch processing time period expires, the CPU transmits the data or execution sequence in the GPU execution cycle I to the GPU for processing; follow steps (1) to (2) to cycle until the application program processing ends. This method utilizes GPU to obtain high throughput while ensuring real-time processing of all network streams.

Description

technical field [0001] The invention relates to the fields of hardware GPU acceleration, high-speed network processing, and real-time scheduling, in particular to a method for processing real-time network stream application programs using GPU acceleration. Background technique [0002] There are many real-time applications on the current network. These real-time applications are related to user experience and have great business needs. Typical applications, such as Internet TV, live video, video conferencing, cloud games, and VoIP, are all related to user privacy, and generally require compression, encryption, and other processing. This poses a challenge to the processing capabilities of network servers and intermediate devices, and causes a large amount of processing overhead. As a result, many operations such as encryption that should protect user privacy have not been processed by service providers, resulting in copyrighted online videos and other content that may be sto...

AI Technical Summary

Problems solved by technology

[0011] Among them, the high latency and execution time uncertainty caused by 1~3 characteristics make GPU not used to handle real-time applications so far

Especially for its batch processing characteristics, a task needs to wait for a batch of tasks to be executed together to fully utilize the GPU, and it takes a certain amount of time to be transmitted to the GPU for processing, so real-time performance is not guaranteed

By reducing the batch size, the delay can be reduced, but the throughput of the GPU will be reduced at the same time; while increasing the batch size, the throughput can be increased, but the delay will also be increased

Images