To split or to group: from divide-and-conquer to sub-task sharing for verifying multiple properties in model checking

Systematic Classification of Attackers via Bounded Model Checking

“…Isolation reduces the SAT formula by dismissing attackers that are outside the COI of the requirement to be verified. Isolation works similarly to COI reduction (see [13,15,7,18,14]), and it transforms Equation 4 into…”

“…The works by Cabodi, Camurati and Quer [15], Cabodi et. al [13], and Cabodi and Nocco [14] present several useful techniques that can be used to improve the performance of model checking when verifying multiple properties, including COI reduction and property clustering. We also mention the work by Goldberg et al [20] where they consider the problem of efficiently checking a set of safety properties P 1 to P k by individually checking each property while assuming that all other properties are valid.…”

Boosting Verification Scalability via Structural Grouping and Semantic Partitioning of Properties

“…The work most similar to ours is a property-clustering procedure based on COI similarity [7,8]. While a similar goal, their solution requires a quadratic number of comparisons between properties, rendering it prohibitively expensive on large testbenches.…”

“…While scalable (near-linear runtime) and able to group properties with 100% affinity, in practice it is desirable to perform additional grouping of properties which have a small tolerable Hamming distance yet are still high-affinity. Again, we stress that a simple procedure of pairwise comparison to check whether properties are within a small tolerance is prohibitively slow in practice, rendering prior techniques as [7,8] unusable in practice. The following algorithms solve this goal of high-affinity group merging, with high scalability and guaranteed grouping quality.…”

“…Despite the prevalence of multi-property testbenches, little research has addressed the problem of optimal grouping or clustering of properties into high-affinity groups. Selective past work [7,8] has experimentally demonstrated that ideal grouping may save substantial verification resource. However, no scalable online property grouping procedure has been provided; this potential was illustrated as a proof-of-concept using computationally-prohibitive offline grouping algorithms with undisclosed runtime.…”