Specification of failure-handling requirements as policy rules on self-adaptive systems

Abstract

Most adaptive systems have compensation mechanisms for recovering from or preventing failures. However, sometimes a compensation is not essential. Hence, diagnosing and compensating each and every one of their failures may be ineffective. Rather than polluting a requirements specification with fine grained definition of failure-handling conditions, this work aims to increase the flexibility of failure handling in self-adaptive systems using tolerance policies. We allow the expression of conditions in which certain failures may be ignored – i.e., conditions on which a failure will not be compensated. Such policies may lead to reduced costs and performance improvement. The FAST framework consists of the definition of a tolerance policy, the mechanisms to evaluate this policy and a tool to aid the creation of policies.