Reliable transaction management in heterogeneous distributed database systems

Abstract

We present a fault-tolerant concurrency control scheme and a transaction-consistent checkpoint scheme in a heterogeneous database system. The fault-tolerant concurrency control scheme guarantees the site autonomy of local database management systems in heterogeneous database systems. To implement the participants of the two-phase commitment protocol for the fault-tolerant concurrency control scheme guarantees the site autonomy of tor on the top of each local database management system. To guarantee complete site autonomy, each local execution monitor sends a commit message for a global subtransaction in its prepared state of the two-phase commitment protocol to the local database managemint system at which the transaction is locally committed. Therefore, local transactions or global transactions can use the data items updated by a global subtransaction which has been locally committed but has not been globally committed. When a global transaction is aborted by the global decision of the coordinator of the two-phase commitment protocol, all transactions which depend on the global subtransaction also must be aborted. Based on the idea of group commit which is used in main memory database systems, the local execution monitor rolls back all the transactions dependent on the global transaction. By the concept of group commit, each local execution monitor delays the sending of the execution result of a global transaction that has been locally committed but has not been globally committed to the user until the transaction is globally committed.