[Paper Review] Data-centric Dynamic Partial Order Reduction

Marek Chalupa, Krishnendu Chatterjee+3 more|arXiv (Cornell University)|Oct 4, 2016

Fuel Cells and Related Materials16 citations

TL;DR

This paper introduces a data-centric dynamic partial-order reduction (DPOR) method based on observation equivalence, where two traces are equivalent if every read event observes the same write event in both. The approach is provably optimal for acyclic architectures, exploring exponentially fewer traces than Mazurkiewicz-based DPOR by using polynomial-time per equivalence class, significantly reducing exploration time and state space size in practice.

ABSTRACT

We present a new dynamic partial-order reduction method for stateless model checking of concurrent programs. A common approach for exploring program behaviors relies on enumerating the traces of the program, without storing the visited states (aka stateless exploration). As the number of distinct traces grows exponentially, dynamic partial-order reduction (DPOR) techniques have been successfully used to partition the space of traces into equivalence classes (Mazurkiewicz partitioning), with the goal of exploring only few representative traces from each class. We introduce a new equivalence on traces under sequential consistency semantics, which we call the observation equivalence. Two traces are observationally equivalent if every read event observes the same write event in both traces. While the traditional Mazurkiewicz equivalence is control-centric, our new definition is data-centric. We show that our observation equivalence is coarser than the Mazurkiewicz equivalence, and in many cases even exponentially coarser. We devise a DPOR exploration of the trace space, called data-centric DPOR, based on the observation equivalence. 1. For acyclic architectures, our algorithm is guaranteed to explore exactly one representative trace from each observation class, while spending polynomial time per class. Hence, our algorithm is optimal wrt the observation equivalence, and in several cases explores exponentially fewer traces than any enumerative method based on the Mazurkiewicz equivalence. 2. For cyclic architectures, we consider an equivalence between traces which is finer than the observation equivalence; but coarser than the Mazurkiewicz equivalence, and in some cases is exponentially coarser. Our data-centric DPOR algorithm remains optimal under this trace equivalence. Finally, we perform a basic experimental comparison between the existing Mazurkiewicz-based DPOR and our data-centric DPOR on a set of academic benchmarks. Our results show a significant reduction in both running time and the number of explored equivalence classes.

Motivation & Objective

• To address the exponential explosion in trace space during stateless model checking of concurrent programs.
• To develop a coarser equivalence relation than the traditional Mazurkiewicz equivalence that preserves verification completeness.
• To design a dynamic partial-order reduction algorithm that is optimal with respect to this new equivalence.
• To achieve significant performance improvements in running time and number of explored equivalence classes compared to existing DPOR methods.

Proposed method

• Proposes a new trace equivalence called observation equivalence, where two traces are equivalent if every read event observes the same write event in both.
• Develops a data-centric DPOR algorithm that explores exactly one representative trace per observation equivalence class for acyclic architectures.
• Uses on-the-fly conflict detection and persistent set techniques adapted to observation equivalence for efficient exploration.
• Introduces a refined equivalence for cyclic architectures that is coarser than Mazurkiewicz but finer than observation equivalence.
• Employs polynomial-time computation per equivalence class, avoiding NP-oracle calls used in prior SMT-based approaches.
• Extends the algorithm to handle cyclic architectures while maintaining optimality under the refined equivalence.

Experimental results

Research questions

• RQ1 Can a coarser equivalence than Mazurkiewicz equivalence be used in dynamic partial-order reduction for concurrent programs?
• RQ2 Is it possible to design a DPOR algorithm that is optimal with respect to a data-centric equivalence relation?
• RQ3 Can such an equivalence relation lead to exponential reduction in the number of explored traces compared to traditional methods?
• RQ4 How does the performance of the new algorithm compare to existing DPOR techniques in practice?
• RQ5 Can the approach be extended to handle cyclic architectures while maintaining efficiency and correctness?

Key findings

• The observation equivalence is coarser than Mazurkiewicz equivalence and in some cases exponentially coarser, enabling significant reduction in the number of traces to explore.
• For acyclic architectures, the data-centric DPOR algorithm explores exactly one representative per observation class and runs in polynomial time per class, making it optimal.
• The algorithm reduces both the number of explored equivalence classes and running time in experiments on academic benchmarks.
• The approach avoids reliance on NP-oracle calls (e.g., SMT solvers), unlike MCR, which suffers from exponential worst-case complexity per class.
• In contrast to MCR, the algorithm guarantees that each equivalence class is explored only once, avoiding exponential blowup in exploration.
• The experimental results show a significant reduction in both running time and number of explored classes compared to traditional Mazurkiewicz-based DPOR.