Load balancing container scheduling method for component dependency

Abstract

The invention discloses a load balancing container scheduling method for component dependency, including the following steps: (1) periodically acquiring resource load information of each host on a host cluster; (2) sending the resource load information of each host to a resource manager to perform unified storage and management; (3) generating a communication matrix; (4) generating a dependency matrix; (5) enabling a container scheduling module to score each host according to the communication matrix, the dependency matrix and the resource load information of each host read from the resource manager, and selecting an optimal host to perform deployment and distribution on container components according to score results; and (6) when new container components need to be distributed, repeatedly executing the steps 1 to 5. The load balancing container scheduling method disclosed by the invention has the advantages that the performance of applications running in all containers reaches overall optimization, the resource load balancing of the host cluster can also be taken into account, and the like.

Description

technical field [0001] The invention relates to a load balancing container scheduling method, in particular to a component dependency-oriented load balancing container scheduling method. Background technique [0002] In recent years, with the promotion and development of the concept of cloud computing, cloud computing has been researched and applied unprecedentedly in both the scientific and industrial circles. The classic cloud computing architecture includes three-tier services of IaaS, PaaS and SaaS. However, with the continuous innovation of computer technology, iterative updates of application versions are becoming more and more frequent. At this time, traditional cloud computing uses virtual machine technology as the smallest granularity resource scheduling unit, which has exposed various problems, such as resource utilization. Low, slow scheduling and distribution, inconsistent software stack environment and other issues. The current popular container technology can ...

AI Technical Summary

Problems solved by technology

In order to further improve the performance of container scheduling, Lin Weiwei's team proposed a Docker cluster container scheduling optimization strategy based on genetic algorithm, using genetic algorithm to search the characteristics of the global solution space, and find the most balanced cluster load in all possible allocation lists. As a result, however, when calculating the individual fitness, the function used to simulate the resource load change when the computing node runs a certain container cannot be completely close to the actual running situation, and needs to be further improved

Jose Monsalve and others proposed a dynamic CPU resource allocation method. The author believes that for some applications, excessive resource usage will seriously affect performance, and extends the concept of time series to the level of virtualized containers. Provide a solution to this problem, but there is a relatively shortcoming that this document aims to study the improvement of CPU resource utilization, and lacks research and improvement on other resource utilization such as memory, which has certain limitations.

[0004] Although a lot of research work has been carried out on container scheduling methods in recent years, the current container scheduling methods all focus on the utilization of various resources to achieve load balancing, while ignoring the resources running in different containers. Applications on the Internet are often associated, especially in the current big data environment. A project often needs to be divided into multiple small application components. These application components are closely related and need to work together to complete tasks. Some methods even The consideration is to improve the utilization rate of a single resource, which has relatively large limitations in practical applications. These methods are not oriented to specific applications, so that relatively good performance effects will not be achieved in actual application scenarios.

Images