可重构系统的演化修复机制

The Evolvable Recovery of Reconfigurable System

利用演化算法实现系统自修复是一种新的容错设计思路,但是演化是一个非常耗时的过程.已有的演化容错系统多属于静态演化,演化过程仅发生在系统设计阶段,系统在运行过程中不具有演化修复的能力.这类演化容错系统虽然可以避免演化耗时,但是只能修复已知错误,无法修复未知错误.针对上述问题,文中提出一种基于动态演化的修复机制,容错系统采用可重构系统和被检测系统的耦合设计方案.当被检测系统出现故障时,可重构系统通过系统演化实现在线自修复.为了减少演化耗时,系统根据错误类型采取不同措施:如果出现已知错误,系统直接在预置配置库中搜索修复配置;如果出现未知错误,则通过动态演化在线生成修复配置,并更新预置配置库.最后,将该容错设计方案用于典型电路的故障模式.实验结果表明,文中提出的演化修复机制提高了系统运行的实时可靠性,预置配置库设计减少了演化耗时.

Utilizing evolutionary algorithms to realize seit-recove~ y l but evolution is time-consuming. Most evolvable fault-tolerant systems so far are off-situation evolution, where the evolution process plays a role iust in the system＇s design stage. In this case, the system does not have capability of evolution recovery during its running time. Although this kind of evolvable fault-tolerant system overcomes the evolution time-consuming, it only repairs the anticipated errors rather than unanticipated errors. To solve the problem, we propose a new recovery approach based on in-situation evolution, the fault-tolerant system is a coupling design of the reconfigurable system and the detected system. Once the target system fails, the reconfigurable system implements in-system evolution to achieve runtime recovery. To decrease the time overhead of evolution process, the proposed system takes different measures according to failure modes： if the anticipated errors are detected, it searches the pre-designed library for recovery configurations; otherwise, when the unanticipated errors happened, it generates new recovery configurations online by evolvable hardware, and updates the pre-designed library by the evolved configurations. At last, this fault-tolerant method is equipped in different failure modes of some typical circuits. The experiments illustrate that the proposed evolvable recovery approach promotes the reliability, and pre-designed configuration library decreases the evolution time overhead.