Design And Implementation Of Docker Swarm Cluster Scheduling Method Based On Improved Artificial Bee Colony Algorithm

The rapid development of cloud computing technology greatly facilitates the network efficiency but at the same time brings new challenges.The most essential problem is the load and resource utilization of the clusters.To solve such resource utilization problem,container technology has been proposed.Compared with that of the traditional virtual machines,the resource occupancy rate of container technology is lower.However,along with the rapid growth of computation demands,the load and resource utilization problem turns out to be also critical in container based system.In fact,containers can be built in batches within the cluster and each node is able to carry a large number of containers.As a result,in the cases that the containers are not uniformly distributed,or the needs of nodes do not match to the number of containers,the cluster resources are unbalanced,leading to a low utilization rate and great resources waste.This thesis focuses on the above mentioned cluster resource allocation problem in the Docker container based system,which is the most popular one nowadays.A scheduling strategy based on improved artificial bee colony algorithm(ABC)has been proposed in this work.The main research work includes the following aspects:Firstly,the existing cluster management tools are analyzed and compared.Among them,Docker Swarm,a native Docker cluster management tool,is selected as the fundamental tool in this research.The structure and internal modules of Docker Swarm is studied in details.Through investigation,the deficiency of existing built-in scheduling strategies is concluded and consequently a new dynamic weighted resource balancing algorithm was proposed.The proposed algorithm takes more comprehensive consideration of the proportion of each resource block and the container's emphasis on resources.And it also takes into account the balance of node resources.Secondly,we apply ABC algorithm to the container cluster scheduling,with the purpose of further improving the performance of the proposed algorithm.The fundamental theories,optimization,application and other aspects of ABC are introduced,with which the proposed algorithm is illustrated in details.The basic idea is to introduce a global optimal solution with ABC,which help improve the algorithm's convergence speed and local search ability.The introduced global optimal solution is the optimal node server,which is obtained by the above mentioned dynamic weighted resource balance algorithm.Finally,in order to collect the node information for calculation,I designed an information acquisition module.This module is able to collect the information such as the CPU utilization,memory utilization,network load,and disk utilization of each node.When the user applies for a new container,this module will send the corresponding node information to the scheduler.The scheduler can then utilize the information to calculate the weights of the nodes and apply the ABC based algorithm for resource scheduling.The performance of the proposed system as well as the ABC based algorithm is tested and evaluated.It is found that the improved scheduling strategy is better than the reference one.In the case of single-container scheduling,the advantage is not obvious.However,in the situation with multi-container scheduling,the improved scheduling strategy can better complete the scheduling work,allocate containers according to the allocated resources to the appropriate node server,and use the resources more efficiently.