Live video cloud transcoding resource allocation and scheduling method based on deep reinforcement learning

Abstract

The invention provides a live video cloud transcoding resource allocation and scheduling method based on deep reinforcement learning, and belongs to the field of machine learning and the field of multimedia content processing. According to the method, in a training stage, a transcoding process is simulated in a simulation environment, a neural network capable of dynamically adjusting cloud computing resource allocation according to the workload change of a transcoding task is trained, and the transcoding task of live video content is scheduled on the allocated cloud computing resource; and inthe execution stage, a real-time decision of resource allocation is made according to the system state by using the trained neural network, and a transcoding task is scheduled on a cloud platform. Based on the deep reinforcement learning method, the cloud computing resource scale can be dynamically adjusted according to the transcoding task load change, and the resource use overhead is saved; based on the estimation method for the upper limit and the lower limit of the transcoding task time, scheduling execution of the transcoding task can be completed in time, and the service quality requirement is met.

Description

technical field [0001] The invention proposes a live video cloud transcoding resource allocation and scheduling method based on deep reinforcement learning, which belongs to the field of machine learning and multimedia content processing. Background technique [0002] In recent years, mobile live broadcast platforms represented by Twitch, Douyu, and Kuaishou have achieved great success in the market. Currently, more than 3.2 million streamers start broadcasting on Twitch each month, with more than 150,000 daily active viewers. On these platforms, the production and production of video content is no longer limited to professional content production organizations, and a large number of ordinary users can also use various terminal devices to share life content such as games and entertainment with global audiences. [0003] As a bridge connecting content producers and content consumers, the live broadcast platform needs to convert the video uploaded by the anchor into multiple ...

AI Technical Summary

Problems solved by technology

In addition, the live broadcast time and duration of the anchor are very random, so the number and length of live streams received by the cloud dynamically change with time

(3) Uncertainty: The content of the live broadcast of the host changes every time. For the cloud platform that provides transcoding services, it is impossible to accurately predict the transcoding time of the current live stream through the historical records of the transcoding time of the same content.

In this method, since the decision-making rules for controlling resource allocation are formulated according to preset scenarios or conditions, when the state of the system changes beyond the preset rules, it is often impossible to make timely adjustments, resulting in over-allocation or under-allocation of resources, affecting The operation of live broadcast business; in terms of transcoding task scheduling, the existing methods mainly adopt the strategy of first-come-first-served, that is, arrange the scheduling priority for the existing transcoding tasks according to the arrival time, and prioritize the execution of the tasks that arrive first. This method does not consider the difference in transcoding time due to different live broadcast content, which may easily lead to transcoding timeout

Images