Applying non-constant volatility analysis methods to software timeliness.

Timing analysis is the application of one or more well-established predictive methods to derive the likely timing behaviour of a specific software task executing on a particular hardware platform. Current approaches towards timing analysis are predicated on the presumption that the software under test is always fixed, i.e., it remains unchanged once deployed to the target hardware. A dynamically adaptable system modifies its behaviour in unanticipated ways, and at unpredictable intervals, to exploit the prevailing operational environment. However, when the software is capable of runtime adaptation, statically derived timing estimates are incapable of accurately capturing the changes in the software timeliness caused by functional adaptations. Traditional timing analysis methods cannot be applied to a dynamically adaptive system, due to the inconstant nature of the software, the unpredictable scheduling of functional adaptations, and the need to produce timing estimates at runtime. This paper describes a work in progress with the aim of statistically forecasting software timeliness using non-constant volatility methods. We outline how timing bounds may be derived for an adaptable software system, with changeable underlying functionality, and show how timing predictions can be modified, using novel statistical models, to mirror runtime functional changes to the software.