The Ariadne Debugger: An Overview



next up previous
Next: Event-Based Behavioral Abstraction Up: Event and State-Based Debugging Previous: Introduction

The Ariadne Debugger: An Overview

 

Ariadne is a post mortem, event-based debugger [16][13]. It matches user-specified models of intended program behavior against actual program behavior as captured in event traces [2]. The traces record low-level interprocess communication, pre-defined language specific-events, and user-defined events. They are stored internally as execution history graphs with nodes representing events and edges representing orderings imposed by Lamport's happened before relation [17]. Ariadne matches user-models against the execution history graph, extracting matched or partially matched subgraphs for further investigation. The extracted subgraphs can be explored through additional modeling, functional queries, and manipulations of logical, temporal orderings.

Ariadne's modeling language for describing intended behavior is quite simple. It was designed to allow efficient matching and precise feedback on errors. Communication patterns are described in terms of a three-level hierarchy of chains, p-chains, and pt-chains with a notation based on regular expressions.

When a pt-chain is matched or partially matched, its events are returned as a match tree in which internal nodes represent user-defined abstractions and leaves represent primitive events. The levels of a match tree correspond to the hierarchical levels of its model.

Ariadne's matching language is quite simple and, as a result, it is not always expressive enough; often its patterns are too coarse, matching program behavior in unexpected ways. To compensate, Ariadne provides a scalable, spread-sheet like interface for exploring match trees, called AVE (Ariadne Visualization Engine) [14]. AVE features the automatic compression of hierarchical visualizations, dynamic attribute calculation, and the partitioning of matched behaviors and attributes. Together these features allow a user to answer questions about a match such as: Did the number of events decrease on each iteration of a loop? Which processes were active during the third iteration? Which processes executed successful comparisons? and Why did the match fail on process 12?

Ariadne was designed for explicitly parallel programming languages in which all interprocess communication occurs through read and write operations. In the next section, we discuss the modifications we made so its modeling language would be appropriate for pC++.



next up previous
Next: Event-Based Behavioral Abstraction Up: Event and State-Based Debugging Previous: Introduction



Sameer Suresh Shende
Fri Apr 5 10:36:18 PST 1996