Optimization Of I/O Isolation For Hybrid Deployed Services

Cloud service clusters are usually configured with the highest requirements of load,which results in the low resource utilization of the clusters for most of the time in actual and serious waste of resources.To increase resource utilization and reduce operational costs,service providers deployed multiple services on the same physical machine by virtualization.The widespread use of container increased scenarios for hybrid deployment,the number and types of services has increased.However,different services have different performance indicators,and it is difficult to ensure that all services are functioning in a unified manner.Secondly,due to the sharing of the resources,the competition of resources are more intense as the number of services increases.In addition,the poor resource isolation between the containers,especially storage resources,which is caused by that the container utilizes the operating system of the host directly to achieve low overhead,will lead the performance of services to be disturbed.Therefore,it is an urgent problem to be solved that how to ensure service performance isolation in complex services hybrid scenarios,which is of great significance to the optimization of cloud service clusters and the development of virtualization technologies.In order to solve the above problems,this dissertation first analyzes the existing public cloud service load,abstracts a typical single-service hybrid scenario,and proposes a hierarchical optimization method of I/O isolation for container;Summarizing the characteristics of multi-services hybrids for the broader generalization scenario,and a load-adaptive optimization method of I/O isolation based on priority is proposed;Finally,the I/O isolation optimization framework FOREST is designed and implemented to optimize the performance of each service in complex hybrid scenario automatically and efficiently.The main research work of this dissertation are as follows:(1)By deeply analyzing the load characteristics and deployment rules of public cloud service clusters,a typical single-service hybrid scenario is proposed,in which is a single latency-sensitive service and a single throughput-first service;According to the service characteristics of the scenario,the optimization method of dual mechanism fusion is studied,which includes that controlling the I/O concurrency to constraint lantency and allocating the disk resources reasonably to ensure the throughput;A hierarchical optimization method of I/O isolation for container based on feedback adjustment is proposed to improve the performance of each service in a single-service hybrid scenario.(2)By analyzing the inter-service performance interference in the generalized multi-service hybrid scenario,that is,the hybrid deployment scenario of multiple delaysensitive services and multiple throughput-priority services,an optimization method which staggered time is proposed;Design service priority auto-division algorithm,and a load-adaptive optimization method of I/O isolation based on priority is studied,which can accurately implement various optimization methods to enhance the I/O isolation of multi-services hybrid scenario under the premise of considering the interaction between similar services and the delay of optimization feedback.(3)Based on the above two I/O isolation optimization methods,the low-cost isolation optimization framework FOREST is designed and implemented,which can determine the current type of service scenario automatically according to the collected service performance data characteristics,switched the optimization method flexibly and calculate the corresponding optimization parameters.FOREST can improve the overall system's adaptability and performance isolation efficiently.In addition,the effects of the individual optimization methods are verified and the potential optimization factors are analyzed,and a large number of experimental results show that FOREST can effectively guarantee the performance of each service in the complex services hybrid scenario.In this dissertation,two isolation optimization methods are designed and a comprehensive I/O isolation optimization framework FOREST is proposed.The experimental results fully demonstrate the effectiveness of the optimization method.The research in this dissertation is of great significance for the analysis and optimization of container I/O isolation in hybrid scenarios,and provides basic theoretical guarantee and technical support for complex scenarios of hybrid deployment of various services.