A Microservice-based Workflow Scheduling Approach With Hybrid Provisioning In Clouds

In recent years,more and more researchers have focused on the microservice.Complex applications are mostly represented in the form of scientific workflows,so using a microservicebased workflow to represent a set of microservice task with partial order relationships is also a useful way for application service providers to build business processes.The rise of container technology provides convenience for the execution and deployment of microservices and improves resource utilization.The cloud service provider provides a variety of resource leasing methods.The hybrid provisioning with on-demand instance and spot instance can greatly reduce the rental cost under the premise of ensuring the reliability of task execution.Therefore,scheduling deadline-constrained microservice-based workflow to containers deployed on virtual machines with hybrid provisioning to minimize the total rental cost,which has important practical significance and application prospects.Compared with other NP-hard workflow scheduling optimization problems,the challenges faced by this problem include two aspects:(1)How to fully utilize the reliability of on-demand instances and the low price of spot instance,to minimize the execution cost of the microservice-based workflow under the premise of meeting deadline constraint.(2)How to efficiently deploy and orchestrate containers on virtual machines under multi-dimension resource constraints to maximize resource utilization.For these challenges,a microservice-based workflow scheduling algorithm based on heuristic rules with hybrid provisioning is proposed in this thesis.The algorithm consists of five components: sub-deadline distribution,ready task sequencing,rule-based task scheduling,online adjustment,and spot instance monitor.The sub-deadline distribution phase proposes three rules to assign sub-deadlines to tasks in the microservice-based workflow.As long as tasks meet their sub-deadline constraints in the scheduling,the deadline constraint of the entire workflow can be guaranteed;In the sequencing stage,three rules are proposed to sort the ready tasks in the ready queue to generate an initial scheduling sequence.The rule-based task scheduling stage searches for available virtual machine instances in the resource pool on the premise of satisfying the task partial order constraint and computing resource constraints for task scheduling and container deployment according to the heuristic rules;the online adjustment phase evaluates whether the successor task of the completed task is a ready task,and adds the ready task to the ready queue.The spot instance monitor is responsible for monitoring the status of the spot instance in real time and taking action to solve out-of-bid failure.In order to evaluate the performance of the proposed algorithm,the multi-factor analysis of variance is used to analyze the relevant parameters and components of the algorithm to obtain the best parameters and determine the algorithm components.Then the proposed algorithm and the benchmark algorithm are compared in different scale workflow instance.The experimental results show that the proposed algorithm outperforms the benchmark algorithm under different deadline constraints.