Efficient Distributed SAT and SAT-Based Distributed Bounded Model Checking

Abstract

Abstract. SAT-based Bounded Model Checking (BMC), though a robust and scalable verification approach, still is computationally intensive, requiring large memory and time. Interestingly, with the recent development of improved SAT solvers, it is frequently the memory limitation of a single server rather than time that becomes a bottleneck for doing deeper BMC search. Distributing computing requirements of BMC over a network of workstations can overcome the memory limitation of a single server, albeit at increased communication cost. In this paper, we present: a) a method for distributed-SAT over a network of workstations using a Master/Client model where each Client worsktation has an exclusive partition of the SAT problem and uses knowledge of partition topology to communicate with other Clients, b) a method for distributing SATbased BMC using the distributed-SAT. For the sake of scalability, at no point in the BMC computation does a single workstation have all the information. We experimented on a network of heterogenous workstations interconnected with a standard Ethernet LAN. To illustrate, on an industrial design with ~13K FFs and ~0.5M gates, the non-disributed BMC on a single workstation (with 4 Gb memory) ran out of memroy after reaching a depth of 120; on the otherhand, our SAT-based distributed BMC over 5 similar workstations was able to go upto 323 steps with a communication overhead of only 30%.

DiVer: SAT-Based Model Checking Platform for Verifying Large Scale Systems

“…DiVer houses the following SAT-based engines, each geared towards verifying large systems: bounded model checking (BMC) [7] and distributed BMC (d-BMC) over a network of workstations (NOW) [12] for falsification, proof-based iterative abstraction (PBIA) for model reduction [13], SAT-based unbounded model checking (UMC) [15] and induction for proofs [5,11], Efficient Memory Modeling (EMM) [14] and its combination with PBIA in BMC for verifying embedded memory systems with multiple memories (with multiple ports and arbitrary initial state) and to discover irrelevant memories and ports for proving property correctness [17].…”

“…For commonly occurring properties, we use customized translations of LTL properties that involve partitioning the problem and using incremental model checking [7]. [12] overcomes the memory limitation of a single server to provide a scalable approach for carrying out deeper search on memory-intensive designs. We achieve a) scalability by not keeping the entire problem data on a single processor, and b) low communication overhead by making each process cognizant of the partition topology while communicating; thereby, reducing the process's receiving buffer with unwanted information.…”

“…Compared to symbolic model checking [1,2] based on Binary Decision Diagrams [3], SAT-based model checking techniques [4][5][6][7][8][9][10][11][12][13][14][15][16][17] have been able to scale and perform well due to the many recent advances in DPLL-style SAT solvers [18][19][20]. We present a SAT-based model checking platform (DiVer) based on robust and scalable algorithms [5-7, 11-17, 20, 21] that are tightly integrated for verifying large scale industry designs.…”

Speculative Image Computation for Distributed Symbolic Reachability Analysis

“…We detail this slicing scheme in Section 6. In addition to these works on symbolic and explicit approaches, reference [15] develops a distributed approach for SAT-based bounded model checking, where the satisfiability (SAT) problem is partitioned onto a NOW and clients can efficiently communicate with each others according to the partition topology.…”

Verification of Embedded Memory Systems using Efficient Memory Modeling

“…To address the memory explosion problem, SAT-based distributed BMC has been proposed [17] in which the BMC problem is partitioned over a network of workstations. However, this technique is not geared towards verifying embedded memory systems.…”