Conflict free transaction scheduling using serialization graph for real-time databases

Abstract

A best effort approach to data scheduling, such as optimistic concurrency control in real-time database systems (RTDBS), imposes a heavy burden on the systems by restarting conflicting transactions. The restarted transactions themselves may miss their deadlines and the resources consumed by them may be wasted. Hence it can be better to schedule transactions such that only conflict free transactions can be executed concurrently at one time. This study explores this approach by making use of serialization graph testing. A serialization graph is used to enforce the serializability of transactions. Only transactions without data conflicts with the executing transactions will be allocated CPU. Consequently, conflict free concurrency among executing transactions can be achieved. All resources including CPU, I/O and data objects will not be wasted on restarted transactions. Therefore, the system can sustain a higher workload. We also devise a real-time serialization graph that considers the timing constraints of transactions. By using our protocols, only a limited amount of transaction delay overhead is observed. However, experimental results confirm that the overall performance of our protocols is better than the real-time optimistic concurrency control (OCC) protocol that is reported as one of the best performing data scheduling approaches in RTDBS.