Input Space Partitioning to Enable Massively Parallel Proof

Osiris

“…Of course, this is not a controller that could be deployed for a commercial automotive system, it is simply for benchmarking of our tool. is relaxed property was provable in 30 seconds, which compares favorably with an analysis time of approximately four hours in [11] on a machine with the same speci cations. e case study computations were carried out on a machine with 44 cores, with available hyperthreading to 88 threads and 256 GB of RAM.…”

“…We have demonstrated the e ectiveness of our approach on a public benchmark, which consists of an adaptive cruise controller with lookup tables. e best published analysis time for this model is around four hours [11], and Osiris has completed the analysis in thirty seconds on a machine with the same speci cations.…”

“…is abstraction is then used as part of an SMT query to check that the system speci cation is satis ed. e work of [11] uses a technique to eliminate proof cases that are not reachable, and then analyzing the remaining cases in parallel with the theorem prover PVS. [11] run their technique on the same public benchmark as we do, and reduce the approximately seven million proof cases to approximately seventy thousand for a runtime of approximately four hours.…”

“…e work of [11] uses a technique to eliminate proof cases that are not reachable, and then analyzing the remaining cases in parallel with the theorem prover PVS. [11] run their technique on the same public benchmark as we do, and reduce the approximately seven million proof cases to approximately seventy thousand for a runtime of approximately four hours. In contrast, our tool uses an SMT backend instead of a theorem prover.…”