马尔可夫决策求解的软件相继故障修复方法

Software Sequential Fault Repair Method for Markov Decision Solving

为提高软件修复效率,减少因软件故障造成的死机和数据丢失问题,提出基于变异技术的软件相继故障自修复方法。选取三种不同类型的变异算子,通过析取范式的方式分析各算子之间的关系,确定变异规则,通过计算怀疑度得出每个变异体的怀疑度值,生成列表,根据排序结果找出故障位置;利用状态空间、动作空间概率分布等元素,建立马尔可夫决策模型,结合修复时间、奖惩值等信息计算修复代价;以累计代价最小为目标,构建决策树,求解马尔可夫决策模型,当累计代价最小时即可输出最优的故障修复策略。实验结果表明,所提方法减少了系统开销,修复效率高,可根据较少的信息执行修复指令,有利于降低系统死机与数据丢失概率。

In order to improve the efficiency of software repair and reduce the problem of crashes and data loss caused by software failures,this paper put forward a self-repairing method for the cascading failure of software based on mutation technology.Firstly,we chose three types of mutation operators,and analyzed the relationship between op-erators by the disjunctive normal form,thus determining the mutation rule.Secondly,we calculated the suspicious de-gree of each variant,and then created a list.According to the sorting results,we found out the location of fault.Moreo-ver,we used the elements such as the probability distribution of state space of action space.Furthermore,we built a Markov decision model to calculate the repair cost based on repair time,and the reward and punishment values.In or-der to minimize the cumulative cost,we constructed a decision tree to solve the Markov decision model.When the cu-mulative cost was minimal,the optimal strategy could be output.Experimental results show that the proposed method not only reduces system overhead,but also has high repair efficiency.In addition,this method can execute repair in-structions with less information,which is beneficial to reducing the probability of system crashes and data loss.