I/O Performance Isolation Optimization On Container-based Clouds

The container technology(such as Docker)is of great value in different kinds of applications,such as continuous testing and deployment,micro-service architecture and traditional cloud services.Comparing with the traditional virtual machine technology,the container technology has several advantages in terms of simplicity,low-overhead,and light-weight,etc.However,as containers share the operating system kernel of the host,the problem of performance interference caused by resources competition between the containers is obvious,that is,the problem of performance isolation is outstanding and it is a challenging problem to guarantee the Service-Level Agreement(SLA)of each container.Hard resources such as CPU and Memory can improve isolation by predicting distribution,due to the I/O load has more real-time,dynamic and other features,it is very difficult to ensure good I/O performance isolation at the same time when the high I/O utilization is guaranteed,currently there is no perfect solution to this problem in the industry.In order to improve the I/O performance isolation of the container-based system,this paper verified and analyzed the problem,presented and demonstrated a feasible solution,the specific work is as follows.First of all,this paper has carried on the deep discussion and the verification of the research content.This paper defined the I/O performance isolation of the container-based system from the perspective of the I/O service's SLA and analyzed the I/O model of container-based systems.Meanwhile,by designing and completing the multi containers I/O load experiments,this paper obtained the variation trend of the I/O performance parameters,the relationship between some parameters,and the result of how an overloaded container influenced on the normal containers.Moreover,based on the experimental results,this paper innovatively proposed a strategy to improve the I/O performance isolation of the container-based system by dynamically restricting the I/O intensity of overloaded containers,and designed a model of system architecture to implement this strategy.Aiming at the algorithm module in this architecture,this paper designed the SLAE algorithm which is responsible for guaranteeing the SLA of I/O services' performance and the UTE algorithm which is used to balance the I/O utilization of the container-based system.Both of these two algorithms can be used to measure I/O performance isolation status of the container-based system in real time,calculate and adjust the I/O load strength of overloaded containers,thereby reducing I/O performance interference on other normal containers.Finally,based on the designed system architecture and algorithms,this paper implemented Seagull,a system for improving I/O performance isolation of Docker container systems.Seagull ensures I/O performance isolation for normal containers by monitoring the I/O performance of the system's containers in real time and dynamically adjusting the I/O load of the overloaded containers by using the SLAE algorithm or the UTE algorithm.Based on the Go language,Seagull supports both console and HTTP API service invocation methods.The results showed that the SLAE algorithm and the UTE algorithm implemented in Seagull in the respective application scenario has in the precondition of maintaining container-based system's high utilization of I/O resources,at the same time to ensure the normal containers' I/O performance threshold,which can improve the I/O performance isolation of the container-based system,it is of great significance for the research and solution of this problem.