[论文解读] On the Workflow Satisfiability Problem with Class-Independent Constraints

Jason Crampton, Andrei Gagarin+3 more|arXiv (Cornell University)|Apr 14, 2015

Access Control and Trust参考文献 9被引用 4

一句话总结

本文引入了类无关（CI）约束，作为用户无关约束的推广，用于建模工作流中的基于组的访问控制。该文提出了一种固定参数可满足性问题（WSP）的固定参数可满足性（FPT）算法，实验表明，该算法在处理复杂实例，尤其是不可满足实例时，优于SAT4J，同时在各种工作负载下保持了高效率和完备性。

ABSTRACT

A workflow specification defines sets of steps and users. An authorization policy determines for each user a subset of steps the user is allowed to perform. Other security requirements, such as separation-of-duty, impose constraints on which subsets of users may perform certain subsets of steps. The \emph{workflow satisfiability problem} (WSP) is the problem of determining whether there exists an assignment of users to workflow steps that satisfies all such authorizations and constraints. An algorithm for solving WSP is important, both as a static analysis tool for workflow specifications, and for the construction of run-time reference monitors for workflow management systems. Given the computational difficulty of WSP, it is important, particularly for the second application, that such algorithms are as efficient as possible. We introduce class-independent constraints, enabling us to model scenarios where the set of users is partitioned into groups, and the identities of the user groups are irrelevant to the satisfaction of the constraint. We prove that solving WSP is fixed-parameter tractable (FPT) for this class of constraints and develop an FPT algorithm that is useful in practice. We compare the performance of the FPT algorithm with that of SAT4J (a pseudo-Boolean SAT solver) in computational experiments, which show that our algorithm significantly outperforms SAT4J for many instances of WSP. User-independent constraints, a large class of constraints including many practical ones, are a special case of class-independent constraints for which WSP was proved to be FPT (Cohen {\em et al.}, J. Artif. Intel. Res. 2014). Thus our results considerably extend our knowledge of the fixed-parameter tractability of WSP.

研究动机与目标

• 为解决涉及用户组而非单个用户的现实世界工作流访问控制需求。
• 将工作流可满足性问题（WSP）的固定参数可满足性（FPT）性质从用户无关（UI）约束推广到更广泛的实用约束类别。
• 设计并实现一种FPT算法，以在实际中高效求解具有类无关（CI）约束的WSP。
• 在大量随机生成的WSP实例上，对所提算法与SAT4J的性能进行评估。

提出的方法

• 引入类无关（CI）约束，其中约束仅依赖于组成员身份，而非个体身份，从而推广用户无关（UI）约束。
• 将先前针对UI约束的工作中的基于模式的方法推广至CI约束，从而实现CI约束下WSP的FPT算法。
• 设计一种基于模式的算法（PBA4CI），系统性地探索在授权策略和CI约束下的有效用户分配。
• 采用一种早期剪枝无效分配的搜索策略，减少搜索空间，同时保证完备性。
• 采用结合模式枚举与约束传播的混合方法，以提升效率。
• 在210个随机生成的WSP实例上进行大量实验，实例中步骤数k ∈ {20,25,30}，约束密度各不相同。

实验结果

研究问题

• RQ1 用户无关约束的概念能否推广以建模工作流中的基于组的访问控制？
• RQ2 当约束为类无关（CI）而非用户无关（UI）时，工作流可满足性问题（WSP）是否仍具有固定参数可满足性（FPT）？
• RQ3 针对实际实例，CI约束下WSP的FPT算法性能与SAT4J相比如何？
• RQ4 所提出的FPT算法在可满足和不可满足实例上是否均保持高效性和完备性？

主要发现

• 所提出的FPT算法（PBA4CI）成功求解了全部210个测试实例，而SAT4J在近40%的实例中失败，主要集中在不可满足实例。
• PBA4CI在97.6%的实例中平均CPU时间优于SAT4J，SAT4J仅在2.4%的实例（共5个）中更快，且这些实例均为约束较轻的。
• 在k=30步时，SAT4J无法对超过65%的实例得出结论，而PBA4CI全部求解成功，表明其在处理难题时具有更强的鲁棒性。
• PBA4CI在不可满足实例上更高效，而SAT4J在可满足实例上表现更优，表明两者搜索策略行为存在差异。
• 随着约束密度增加，算法性能提升，且在较大k值（最高达30步）下仍能有效扩展。
• FPT算法已公开发布，附带实验数据，支持可复现性及进一步基准测试。