Research And Implementation Of Strategy Effectiveness Guarantee Mechanism For Self-Adaptive Software

The changing environment and increasingly complex structure of software system make the maintenance of software more and more difficult.Many scholars began to study Self-Adaptive Systems(SASs)which can adjust themselves to meet functional and non-functional requirements and adapt to the changes of software.By analyzing the environment information,the SASs can generate the adaptive strategy independently.Then the strategy will guide the system to adjust.The dynamic environment brings many problems.It is difficult for the system to obtain accurate environmental information.There will also be some simplifications in the assumptions put forward by the decision-making methods of adaptive software.These will cause the adaptive strategy to deviate from the target,that is,the software requirements.And as software becomes more and more important in various systems,the consequences will be serious once its self-adaptive adjustment is unreliable.Therefore,SASs should ensure that self-adaptive strategies can achieve desired goals.Aiming at this problem,this paper establishes a complete mechanism to ensure the effectiveness of the strategy based on the Partial Observable Markov Decision Process.It supports self-adaptive software to evaluate the effectiveness of strategies during the operation phase and predict the execution effect of self-adaptive strategies.At the same time,it can make online corrections to the failure strategies to ensure the effectiveness of them.This paper consists of four parts.First,proposed method builds a system state model based on POMDP.The modeling method refers to the hierarchical analysis method to determine the weights of multiple related attributes of the software requirements.This links the system state to the adjustment goals of the software.Then it extracts the change relationship of the system state by discretizing and fitting the system log.The system state model based on POMDP not only supports a small amount of state information collected at runtime to describe the system state of the SAS,but also can perform a variety of possible system state inferences to better describe the system state changes in an uncertain environment.The model provides the basis for a strategy effectiveness guarantee mechanism.Secondly,this paper proposes a method for evaluating the effectiveness of strategies.This article describes the relevant attributes of adaptive strategies in the form of multivariate groups.Based on the action sequence and their interactions recorded in the strategy,state inference is performed based on the system state model during the operation phase,with the system state triggering software changes as the starting point.The method predicts the system state that the strategy can reach and evaluates the state to verify the execution effect of the adaptive strategy.Then,in order to respond to software changes as quickly as possible,this paper proposes a strategy correction method based on reinforcement learning to correct failure strategies.This method processes the strategy that cannot reach the expected self-adaptive adjustment target,and obtains the strategy that can achieve the target.Reinforcement learning algorithms take feedback on the effect of adaptive software adjustment targets after strategy execution as feedback,and accumulate reward values for single-step actions in different system states.The revised method traces back in the reverse order of state reasoning during strategy evaluation,backtracking one action at a time,and performing single-step reinforcement learning in a semi-random exploration manner until the modified action sequence meets the needs of adaptive adjustment.Finally,this paper verifies the proposed method through typical e-commerce system cases.Taking common response timeouts of web systems as an example,this article designs an experimental scenario and verifies that the effectiveness guarantee mechanism of the strategy can well evaluate and modify the strategy.This paper analyzes the algorithm time-consuming under different action sequence lengths,and verifies that the performance of the strategy's effectiveness guarantee mechanism is sufficient to support the calculation in the operation phase.