Concurrency control is fundamental to maintaining data consistency in modern database systems. While relational databases commonly employ lock-based mechanisms like Two-Phase Locking (2PL), alternative approaches, such as the lock-free Two-Phase-Ordering (2PO) protocol, have emerged for distributed object storage systems.

This thesis investigates the applicability of 2PO as an underlying concurrency control mechanism for Structured Query Language databases. We establish a system model and explore various implementations, starting with tables as resources and refining to row-based granularity. Our findings indicate that while 2PO can achieve certain isolation levels, its effectiveness is challenged by the dynamic evolution of transaction dependencies.

We demonstrate that with modest extensions, including a “Hunger Phase” for re-evaluating predicates and the integration of Two-Phase-Commit, we can attain Read Uncommitted Isolation and Read Committed Isolation, respectively. However, propagating every state change to continuously update the dependency graph proves impractical, preventing a direct, scalable procedure for guaranteeing fully conflict-serializable outcomes. Our work suggests that achieving serializability in a pessimistic, non-locking manner, especially when resources are unknown or dynamically changing, remains a challenge, with traditional locking mechanisms like 2PL often presenting the most practical solution.