Distributed Dynamic Partial Order Reduction Based Verification of Threaded Software

Abstract

Abstract. Runtime (dynamic) model checking is a promising verification methodology for real-world threaded software because of its many features, the prominent ones being: (i) it avoids the need to extract a model and instead runs the actual code, and (ii) the precision of information available at run-time allows techniques such as dynamic partial order reduction (DPOR) [1] to dramatically cut down the number of interleavings examined. Unfortunately, DPOR does not have many implementations for real thread libraries such as POSIX Pthreads, and suffers from high computational overheads due to a stateless search that requires re-executions. In our previous work [2], we designed a runtime model checker, inspect, that overcomes the first of these drawbacks. Inspect has been shown capable of detecting data races, deadlocks and other incorrect API usages in real-world PThreads C programs. In this paper, we describe a distributed version of inspect, which implements an extended DPOR algorithm. Our two key contributions are: (i) a practical algorithm for distributed dynamic partial order reduction; (ii) the innovations that helped distributed inspect attain nearly linear (with respect to the number of CPUs) speedup on realistic examples.

Scalable Dynamic Partial Order Reduction

“…Note that our presentation of [8] omits the use of sleep sets [7]. This simplification is made to achieve consistency with other techniques [5,21] presented later in this section.…”

“…However, as pointed out by Yang et al [21], such a parallelization suffers from two problems. First, due to the non-local nature in which DPOR updates the exploration frontier, different workers may end up exploring identical parts of the state space.…”

“…To push the limits of systematic testing, existing tools combine of stateless exploration [8] with state space reduction [5,7,9] and parallel processing [21].…”

“…Unlike previous work [21], our approach is based on a novel exploration algorithm that 1) enables trading space complexity for parallelism, 2) achieves load-balancing through time-slicing, 3) provides for fault tolerance, which we consider a mandatory aspect of scalability, 4) scales to more than a thousand parallel workers, and 5) is guaranteed to avoid redundant exploration of overlapping portions of the state space.…”

“…In this section we give an overview of stateless exploration [8], dynamic partial order reduction (DPOR) [5], and distributed DPOR [21], which represent the state of the art in scalable systematic testing of concurrent programs.…”