Polynomial-Time Verification and Testing of Implementations of the Snapshot Data Structure

Abstract

We analyze correctness of implementations of the snapshot data structure in terms of linearizability. We show that such implementations can be verified in polynomial time. Additionally, we identify a set of representative executions for testing and show that the correctness of each of these executions can be validated in linear time. These results present a significant speedup considering that verifying linearizability of implementations of concurrent data structures, in general, is EXPSPACE-complete in the number of program-states, and testing linearizability is NP-complete in the length of the tested execution. The crux of our approach is identifying a class of executions, which we call simple , such that a snapshot implementation is linearizable if and only if all of its simple executions are linearizable. We then divide all possible non-linearizable simple executions into three categories and construct a small automaton that recognizes each category. We describe two implementations (one for verification and one for testing) of an automata-based approach that we develop based on this result and an evaluation that demonstrates significant improvements over existing tools. For verification, we show that restricting a state-of-the-art tool to analyzing only simple executions saves resources and allows the analysis of more complex cases. Specifically, restricting attention to simple executions finds bugs in 27 instances, whereas, without this restriction, we were only able to find 14 of the 30 bugs in the instances we examined. We also show that our technique accelerates testing performance significantly. Specifically, our implementation solves the complete set of 900 problems we generated, whereas the state-of-the-art linearizability testing tool solves only 554 problems.