On-line recovery for rediscovered software problems

This paper discusses a method that can allow a system to avoid or recover from the exercise of certain known faults at runtime and thus make certain software upgrades unnecessary. The method uses the knowledge of the characteristic symptoms of a software fault and the appropriate recovery action for the fault to detect and recover from the future exercise of the fault. An analysis of field data shows that the method is applicable to about 25% of faults and there is a potential for this number to go up. Using the method can not only improve the availability of user applications, but can also reduce the number of rediscovered problems and hence reduce the resources required for software service.