How to Manage Replicated Real-Time Databases in an Overloaded Distributed System ?

In order to meet their temporal constraints, current distributed applications such as Web-based services and electronic commerce use the technique of data replication. To take the replication benefit, we need to develop concurrency control mechanisms with high performance even when the distributed system is overloaded. In this paper, we present a protocol that uses a new notion called importance value which is associated with each realtime transaction. Under conditions of overload, this value is used to select the most important transactions with respect to the application transactions in order to pursue their execution ; the other transactions are aborted. Our protocol RCCOS (Replica Concurrency-Control for Overloaded Systems) augments the protocol MIRROR, a concurrency control protocol designed for firm-deadline applications operating on replicated real-time databases in order to manage efficiently transactions when the distributed system is overloaded. A platform is currently being developped to measure the number of transactions that meet their deadlines when the processor load is controlled. Key-words: Firm real-time transactions, distributed database system, commit processing, processor overload, data replication.