Topological Response to Deadlock Detection and Resolution in Real-Time Database Systems

In distributed environments with shared resources, deadlocks are imminent. In many cases, deadlock detection and resolution incur unacceptable overhead and systems resort to simple timeout mechanisms. This paper demonstrates that there are scenarios where timely handling of deadlocks using efficient protocols can result in enhanced overall performance. At the same time, the network topology plays a significant role and incurs varying degree of overhead depending upon the underlying system configuration. Congestion has been simulated via combination of parameters including workload, arrival rate and update percentage. The key performance measure is the completion rate of transactions before their deadlines as determined from temporal constraints. Since data is partitioned across several nodes, transactions may execute on more than one node by creating sub-transactions. This leads to both local and global deadlocks which are then handled using the proposed protocols.