Partial-Order Reduction for Schedule-Abstraction-based Response-Time Analyses of Non-Preemptive Tasks

Compositional verification of embedded real-time systems

“…For the automotive industrial challenge, our UPPAAL model proved resilient to large clock values, with timing constraints varying from 0 to 1 billion time units. The authors of [73] verified the same case study (also on core 2) but their approach does not support precedence constraints between runnables (i.e., the successor relation that we have in our FSM, Sect. 3.1).…”

“…Therefore, we can provide an indicative, preliminary comparison, only for the optimistic WCRT computations, under the above equivalence assumption and in the absence of release jitters. We notice that our UPPAAL model, albeit evaluated on a mid-range computer that is much less powerful than the one used in [73], is seven times faster than the Scheduling Abstract Graph (SAG) approach [73, Fig. 4(h), Runnable-level, No jitters], and provides exact results contrary to the SAG-POR extension (optimistic WCRTs are exact under the assumption of total absence of overheads).…”

“…schedulable) with the former (resp. the latter), as shown in [73]. As explained at the beginning of this section, we fully automatized Step 0 for this case study as well: a waters_c.xta is automatically generated from the XML description for each core c, given the affinity provided within the same description.…”

“…However, digraphs are destined to uniprocessor problems. The Scheduling Abstract Graph (SAG) [64] and its extensions [65,73] is another line of related work. It leverages techniques known in model checking (state-space construction, partial order reduction) to devise algorithms specific to the schedulability property, making SAG and its extensions a state-of-the-art family of techniques suitable for exhaustive schedulability analysis, especially in the presence of jitters and offsets.…”

“…However, contrary to our approach, the exact version of SAG (i.e., providing exact WCRTs without data sharing, corresponding to our optimistic WCRTs in Sect. 7) is restricted to non-preemptive jobs on uniprocessor platforms [64], which is also the case for the more recent, non-exact POR extension [73]. The multiprocessor version is reserved for global scheduling, and is either non exact with independent tasks (in the limited preemtive setting [65]), or fully non-preemptive with FIFO-spinlock-based data sharing [68].…”