Data center scheduling system and method

Abstract

The invention discloses a data center scheduling system and method. The scheduling system can determine a moment and path allocation scheme by generating conflict-free path sets of a fat tree to ensure the unity of moment allocation and path allocation. The method comprises the steps: information interaction between a center controller and a data layer and a scheduling strategy of the center controller, wherein the center controller needs to collect virtual queue length information from servers, determines a scheduling scheme according to the virtual queue length information and a randomly-selected conflict-free data set and then feeds back scheduling scheme information to the servers and uplink switches. The designed scheduling system only needs the virtual queue length of a current moment and information of the conflict-free path set used at a previous moment, the conflict-free path set used at the current moment is generated by randomly replacing part of paths, and in-network transmission zero delay can be achieved; and an adopted path replacement strategy is only related to the virtual queue length, and the optimal throughput can be achieved by reasonably designing the replacement probability.

Description

technical field [0001] The present invention relates to the field of data center networks, and more specifically, to a data center scheduling system and method. Background technique [0002] With the rapid development of emerging technologies such as cloud computing, big data, and virtualization, data center traffic and bandwidth have grown exponentially. Cisco's latest report shows that cloud data center traffic will exceed 14ZB in 2020, an increase of 262% compared to 2015. Behind the explosive growth of data volume and computing volume is the rapid growth of data storage and computing costs, which promotes the evolution of data centers from server rooms to ultra-large-scale deployments, and the number of data centers with hundreds of thousands or even millions of servers continues to increase. [0003] For ultra-large-scale data centers, it is necessary to provide efficient interconnection among huge server clusters. However, the traditional three-tier data center netwo...

AI Technical Summary

Problems solved by technology

[0006] Aiming at the defects of the prior art, the purpose of the present invention is to solve the deficiencies of the existing data center network scheduling strategy. For example, if the data flow scheduling is improper based on the fat tree topology, some paths will be seriously congested, and the scheduling complexity is extremely high, and the computing resources The consumption is too large. Among the scheduling schemes based on network boundaries, the Fastpass scheme has a high computational cost. Once the data demand prediction of the Mordia scheme is inaccurate, it will lead to technical problems of greater delay.

Images