AI computing power pool cluster management method based on MQTT and HTTP

Abstract

The invention relates to an AI computing power pool cluster management method based on MQTT and HTTP, and the method comprises the steps: S1, calculating the remaining computing power AIk of each nodeaccording to the number of tasks already operated on each node, 0 < k < = n, and k and n are positive integers; S2, storing all Alk into a linked list through a preset strategy; and S3, when there isa calculation task, calling a certain calculation force Alm in the linked list, if the calculation force Alm meets the requirement of the calculation force of the task, scheduling the task to a nodem corresponding to the calculation force Alm for operation, recalculating the residual calculation force Al'm of the node m, and storing the residual calculation force Al'm in the linked list again according to a preset strategy. According to the method, optimal management of the computing power of the cluster equipment can be realized, and compared with the existing single embedded AI chip, a plurality of embedded AI chips can be connected in parallel, so that the high computing power and the low price are achieved through the unified management of the network cluster, and the method is a better solution.

Description

technical field [0001] The present invention relates to the management technology of AI computing power, in particular to a cluster management method of AI computing power pool based on MQTT and HTTP, which is used for the management of AI chip computing power. Background technique [0002] In recent years, with the continuous development and improvement of the AI ​​industry, the continuous penetration of AI applications into pan-security and anti-AI scenarios, and the widespread demand for AI applications from the public, the demand for providing a cost-effective and powerful AI device is also increasing. strong. Although the solution based on the traditional X86 CPU+GPU architecture can meet the computing power requirements of many AI scenarios, the high price prevents the implementation of AI applications in many civilian scenarios. With the development of embedded AI chips, the computing power of a single embedded AI chip is getting higher and higher, but based on AI ap...

AI Technical Summary

Problems solved by technology

Although the solution based on the traditional X86 CPU+GPU architecture can meet the computing power requirements of many AI scenarios, the high price prevents the implementation of AI applications in many civilian scenarios.

With the development of embedded AI chips, the computing power of a single embedded AI chip is getting higher and higher, but based on AI applications in larger scenarios, the computing power of a single embedded AI chip is still slightly insufficient

Images