On uses of extreme value theory fit for industrial-quality WCET analysis

Abstract

Abstract-Over the last few years, considerable interest has arisen in measurement-based probabilistic timing analysis. The term MBPTA has been used to indistinctly refer to a variety of different applications of Extreme Value Theory (EVT) to the timing analysis problem. The successful application of MBPTA techniques to a score of case studies has not fully dispelled the concerns that industrial stakeholders had with the quality of the computed bounds, hence ultimately with their industrial viability. Placing focus on the MBPTA methods and techniques developed in the PROARTIS and PROXIMA projects, collectively referred to as proMBPTA, we discuss the main misconceptions and pitfalls that can prevent a sound application of EVT-based WCET analysis. Using a combination of arguments and support examples, we show that proMBPTA is a rigorous process, fully amenable to sound and sustainable industrial use.

Boosting Guaranteed Performance in Wormhole NoCs with Probabilistic Timing Analysis

“…Some authors point out that exponential tails may be optimistic in some scenarios where either measurements from multiple paths are placed in a single sample or programs with unbounded execution time are analyzed [13]. In our case, as shown in [15], programs have finite execution time and paths are analyzed separately since otherwise the application of MBPTA could be unreliable.…”

Probabilistic Worst-Case Timing Analysis

“…[102] shows that proceeding in that way does not work in all cases. Milutinovic et al [91] discuss the risk of the single-bucket approach, which causes the pWCET to reflect the path traversal frequency as occurred during analysis, which the user may have great difficulty at correlating with the operational behaviour of the program. Authors further show that, when using Gumbel distributions, the single-bucket approach may lead to pWCET estimates that either are optimistic (i.e.…”

Measurement-based cache representativeness on multipath programs

“…Statistical aspects in the application of EVT are explored in [13,32] where EVT parameters selection and con dence intervals are the main concerns. A more informed application of EVT to the so ware domain is advocated instead in [23], restricting EVT to using exponential tail distribution [2] on the account that it has been argued to be the most stable (and always overapproximating) distribution to model worst-case execution of real-time so ware [2,27]. A universally recognized issue in the application of EVT is related to identifying an exhaustively representative input sample [13,32,38]: our work directly responds to this challenge by synthetically deriving a path-upperbounding version of the target program and de ning the minimum number of observations to be collected on such version.…”