STO Verification

Transactional memory is an appealing alternative to lock-based synchronization mechanisms. It provides in-memory operations with transaction abstraction analogous to database systems. Despite recent works on exploring different implementation strategies of software transactional memory, little attention has been paid to verify their correctness. In this work, We aim to verify in Coq the correctness of a particular software transactional memory system, namely STO. While opacity is deemed as a better candidate of correctness criterion for transactional memory, for the purpose of this project, we only verify that STO’s implementation guarantees serializability, a weaker correctness criterion than opacity. We prove that all execution traces generated by STO’s commit protocol are serializable. That is, there is always an equivalent serial trace whose execution result is the same as that of the original trace. Currently, our problem is that our Coq inductive types and definitions become increasingly complex as we strictly follow STO’s commit protocol. We are concerned that usch complexity might indicate a bad design and thus leads to intractable proofs later on. While we strive to prove the serializability theorem, we might end up turning to simpler theorems (e.g., deadlock-free), if the serializability proof cannot be constructed.