Consistency management in distributed heterogeneous information sources
by Lyman Siu-fung Do
THESIS
1996
Ph.D. Computer Science
xiv, 218 leaves : ill. ; 30 cm
Abstract
Accessing scattered information in a network of multiple, pre-existing heterogeneous, and autonomous data sources is a dominant factor in an enterprise's competitiveness. Several advanced transaction models and multidatabase architectures have been proposed to manage the consistency of information in such environments. However, the role of pre-existing local transactions (applications), especially those which update data that has some interrelated counterparts in another information source, is unclear. The update of information by local transactions is either restricted or ignored. When ignored, local updates will violate consistency constraints and result in data inconsistency....[ Read more ]
Accessing scattered information in a network of multiple, pre-existing heterogeneous, and autonomous data sources is a dominant factor in an enterprise's competitiveness. Several advanced transaction models and multidatabase architectures have been proposed to manage the consistency of information in such environments. However, the role of pre-existing local transactions (applications), especially those which update data that has some interrelated counterparts in another information source, is unclear. The update of information by local transactions is either restricted or ignored. When ignored, local updates will violate consistency constraints and result in data inconsistency.
We have developed a new multidatabase framework FlexIS to reason about the behavior of local update transactions in multidatabase environments. Specifically, FlexIS allows concurrent execution of global transactions, local transactions, and most important, a new class of transactions, local asynchronous update transactions (LAUs), identified to handle local updates on globally interrelated data. LAUs are pre-existing local transactions that update globally interrelated information. Under the FlexIS framework, we can (1) reason about the correct behavior of LAUs formally, and (2) manage LAU execution together with global transactions and regular local transactions under different transactional correctness criteria; specifically, we have modeled Global Serializability and Epsilon Serializability.
We also define a model and associated system components, called the Consistency Management Facilities (CMF), which maintain the transaction relationships during the execution of transactions identified in the FlexIS and transactions generated for data consistency enforcement. The CMF support a formal framework to specify the relationship between a transaction which violates an interdatabase consistency constraint and constraint enforcement transactions designed to maintain the consistency of the information. In addition, the CMF define a software architecture and modular interfaces to implement the specified transactional relationships. The formal specification supported in the CMF is portable to various types of database management architectures, from the centralized homogeneous DBMSs to distributed heterogeneous DBMSs.
The transaction management functions in the CMF are inspired from the concurrency control algorithms designed in the FlexIS framework. These algorithms, and the more general transaction management functions in the CMF, can be used to manage asynchronous transactions to achieve a better transaction throughput and yet guarantee information consistency. The FlexIS framework with the Consistency Management Facilities can be used to maintain consistency of loosely coupled information sources for applications which have complex information access and update requirements, such as the migration of legacy systems, or as a backend component of a workflow management system that operates over distributed information sources.
Post a Comment