Analysis of Commutativity with State-Chart Graph Representation of Concurrent Programs
share
We present a new approach to check for commutativity in concurrent programs from their state-chart graphs. A set of operations are commutative if changing the order of their execution on an object does not affect the abstract state of the object and returns the same response. Concurrent operations that commute at object-level can be executed concurrently at transaction-level, which boosts performance while preserving the appearance of atomicity and isolation. Utilizing object-level commutativity in transactional execution enables the reuse of existing non-blocking programming techniques for thread-level synchronization. In our approach, we generate state-chart graphs by tracking data on the atomic instructions invoked on the concurrent object during model checking and represent the atomic instructions as states in a state-transition representation. Considering the non-deterministic nature of concurrent programs, we determine commutativity by exhaustively searching for identical object states captured at a thread-level granularity across all thread interleavings. With this methodology, a user can not only verify commutativity among operations, but also can visually check ways in which methods commute at object-level, which is an edge over current state-of-the-art tools. The object-level commutative information helps in identifying faulty implementations and performance improvement considerations. We use the graph database, Neo4j, to represent object states as nodes that further assists the user to check for concurrency properties using Cypher queries.
READ FULL TEXT