Parallel propositional satisfiability checking with distributed dynamic learning

Partitioning SAT Instances for Distributed Solving

“…, φ n in parallel. The plain partitioning approach bears close resemblance to the guiding path approach [9,15], and, provided that the run time reduction caused by the partition function depends naturally on n, the plain partitioning approach can solve arbitrarily difficult problems given a sufficiently large n [12]. However, plain partitioning has some fundamental flaws, which prevent it from being a practical solving method in distributed environments as such.…”

Multithreaded SAT Solving

“…Furthermore, the solver must still maintain good single threaded performance with these parts included, otherwise the speedup achieved through the use of multiple processors might be negated. For example, PaSAT achieved an excellent speedup of 40 on a benchmark set called longmult using 24 processors [18]. However, the single threaded solver took thousands of seconds to solve each instance in the benchmark set.…”

“…This system also reduces the latency within the clause sharing system. Note, PaSAT later combined Message Passing and the "Clause Store" in [18] to allow better system scaling.…”

“…This design takes full advantage of the low latency and bandwidth a shared memory database provides. PaSAT did something similar in [18] with mobile agents that contain thread specific information. This information however, was incomplete and not always up-to-date.…”

“…Work has been done on parallelizing SAT solvers for use on asynchronous distributed systems, using some form of message passing. Such examples are GridSAT [19], PaSAT [21], PaMIRA [23], and others [16,17,18,20,22,24]. Message passing, however, is slow and requires a lot of overhead when compared to a well designed shared memory system.…”