Fitting processor architectures for measurement-based probabilistic timing analysis

Abstract

The pressing market demand for competitive performance/cost ratios compels Critical Real-Time Embedded Systems industry to employ feature-rich hardware. The ensuing rise in hardware complexity however makes worst-case execution time (WCET) analysis of software programs - which is often required, especially for programs at the highest levels of integrity - an even harder challenge. State-of-the-art WCET analysis techniques are hampered by the soaring cost and complexity of obtaining accurate knowledge of the internal operation of advanced processors and the difficulty of relating data obtained from measurement observations with reliable worst-case behaviour. This frustrating conundrum calls for novel solutions, with low intrusiveness on development practice. Measurement-Based Probabilistic Timing Analysis (MBPTA) techniques offer the opportunity to simultaneously reduce the cost of acquiring the knowledge needed for computing reliable WCET bounds and gain increased confidence in the representativeness of measurement observations. This paper describes the changes required in the design of several high-performance features - massively used in modern processors - to meet MBPTA requirements.This work has received funding from the European Community's Seventh\ud 1025 Framework Programme [FP7/2007-2013] under grant agreement 611085 (PROXIMA,\ud www.proxima-project.eu). Support was also provided by the Ministry of Science and Technology of Spain under contract TIN2015-65316-P and the HiPEAC Network of Excellence. Leonidas Kosmidis is funded by the Spanish Ministry of Education under FPU grant AP2010-4208. Jaume Abella has been\ud 1030 partially supported by the MINECO under Ramon y Cajal postdoctoral fellowship number RYC-2013-14717. The authors wish to acknowledge Michael Houston, Liliana Cucu-Grosjean and Luca Santinelli for contributing to the genesis of this work.Peer ReviewedPostprint (author's final draft

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

“…To this end, they define the notion of MBPTA compliance [17], which reflects certain abilities of the observation process and specific features of the execution platform. The latter in particular require applying combinations of (hardware or software) randomization or upper-bounding to the executiontime behaviour of the hardware resources with the highest jitter.…”

“…Applying time randomization ensures that what is observed at analysis is an equal representation of what can happen at operation. Time randomization can be realized directly in the hardware or obtained by software manipulations neutral to functional behaviour on top of COTS processors [8], [12], [17].…”

“…A case for Randomization: Representativeness of the diverse SoJ can be attained by injecting randomization in the timing behaviour of selected hardware or software components [5], [17]. Randomization ensures that several of the sources of execution-time variability are transparently captured in the analysis results without needed direct user intervention (whether by providing specific input vectors or setting the internal state).…”

“…Each, conveniently collected (see [5], [17]), execution-time measurement corresponds to an independent and identically distributed (i.i.d.) observation of a random variable.…”

“…Randomization guarantees that the i.i.d. requirement is satisfied by construction [17] and allows probabilistic reasoning on the representativeness of input data and the quality of results [9].…”