Value-based software metrics

Many software metrics, particularly dependability metrics, are usually presented in a value-neutral framework. For example, the primary metric for Reliability, Mean Time Between Failures, is referenced to a one-size-fits-all (and often not explicitly defined) definition of “failure.” This can lead to serious problems, as when operators’ optimization on liveness produces user response times to queries of up to 2 weeks (an actual example). Clearly, unacceptable response time is a “failure” for some stakeholders that should be reflected in the metric definition. The paper presents some initial thoughts and a candidate approach for addressing this issue. 1. Current Interests USC’s Center for Software Engineering has been working on value-based software engineering for several years; see next section.