A Fault-tolerant Bottlenecks Based Method For Reliability Evaluation Of Cloud Platforms

In the information age,software architecture is gradually changing from single architecture to distributed architecture.This trend also benefits from the infrastructure development of cloud platforms,and the reliability of cloud platform is becoming an important factor restricting the quality and safety of distributed software.To improve the stability and reliability of the cloud platform in real situations,cloud platform usually adopt fault-tolerant design to ensure that the system can still operate normally in case of sudden failure.However,the fault-tolerant capability of any system design is limited,and the reliability of different systems is different.Testers often use fault injection techniques to expose fault-tolerance limitations and fault-tolerant design flaws in cloud platform systems.Most of the traditional fault injection technologies have the problems of high complexity,poor adequacy and low degree of automation.In recent years,the lineage-driven fault injection method has been proposed.By injecting simulated faults into the cloud platform system,the fault combination that can destroy the normal execution of the system can be detected,so as to expose the defects of the fault-tolerant design of the system and evaluate the faulttolerant performance of the system.The existing fault injection methods are generally based on the system topology structure to solve the fault combination that can destroy the normal execution of the system,which has a big limitation on the performance and availability of the algorithm.Secondly,there is a lack of systematic research on the relationship between the fault combination and system reliability.To accurately assess the fault-tolerant performance and reliability of the cloud platform system,and further optimize the fault tolerancedesign,this paper focuses on the concept of minimal fault combination(fault-tolerant bottleneck)that can destroy the normal operation of the system,discusses the research significance of low dimensional fault-tolerant bottleneck,and studies the efficient detection method of low dimensional fault tolerance bottleneck under the cloud platform system structure unknown scenario.On the basis,the method of reliability calculation and weak host location based on fault-tolerant bottleneck is proposed.The specific research contents include:· To solve the problem that the existing fault injection methods have high complexity of detecting system fault combinations and are not suitable when the system topology is unknown,this paper proposes a new low dimensional fault-tolerant bottleneck detection method based on business tracking.This method can be applied to the scenarios of system architecture is unknown and detect the fault-tolerant bottlenecks in the system efficiently through the business tracking technology.In addition,the experimental evaluation module studies the influence of test case execution order on algorithm execution cost,and discusses the application scenarios of incremental algorithm framework.· This paper measures and compares the the reliability of cloud platform systems using the reliability engineering indexes(such as reliability function,mean fail-free time,etc.).To solve the problem that the existing methods are difficult to be applied directly due to the complex business invocation relationship of cloud platforms,this paper proposes a cloud platform system reliability calculation method based on fault-tolerant bottleneck.And two approximate calculation schemes are proposed to solve the problem that high complexity of algorithm leads to poor practicability.· This paper studies the definition of weak host based on fault-tolerant bottleneck and the positioning method of weak host,and studies the significance of weak host to system reliability improvement.This method helps the testers find out the hidden trouble and defects in the system fault tolerance design,locate the key host of the system,and further optimize the system fault tolerance design.