TRANSACTION MANAGEMENT FOR GLOBAL SERIALIZABILITY AND LOCAL-AUTONOMY IN MULTIDATABASE SYSTEMS

Global serializability and local autonomy in the MDBS (multidatabase system) are important issues since MDBS has been studied for the past five years. This paper proposes the transaction processing model for an MDBS and its three-level scheduling algorithm for global schedulers, for integrated local and global transaction schedulers (ILGSs), and for local schedulers to achieve global serializability, local autonomy, and higher concurrency. The most difficult problem to ensure global serializability for an MDBS is to schedule global transactions so that their execution order and their serialization order at each site are identical. In our algorithm, this problem is resolved by two different ILGS modules: the first ILGS module is used in the case that local concurrency control (LCC) algorithm produces a history in which the serialization point of a transaction is the same as its serializable order, and the second ILGS module is used for each LCC algorithm that generates a serializable history. To achieve a high degree of concurrency, transactions are scheduled at the ILGS module, regardless of they are local or global, in a way that unnecessary abort or delay is not caused. By simply adding the ILGS level on top of each local DBMS, local DBMSs need not be changed at all. Therefore, local autonomy is guaranteed. Our algorithm also provides freedom from a deadlock at the global level and may be used even if data items are replicated at many sites.