Recovery in parallel database systems

The success of relational DBMS technology in the commercial market has led to its application in environments demanding high transaction rates, real-time responses, and continuous availability. New multi-processor hardware architectures enable these requirements to be met. This book systematically analyzes the main recovery approaches for centralized DBMSs developed over the last two decades, focusing on their effectiveness in ensuring availability and soft real-time response. The analysis specifically addresses methods used in current commercial and research systems. A notable gap in existing techniques is the inability to automatically and non-blockingly re-establish fault tolerance levels online. The book introduces a set of innovative recovery methods for parallel DBMSs based on multi-computer shared-nothing hardware, designed to support continuously available transaction services, handle very high transaction loads, and achieve soft real-time transaction responses. It provides a well-structured insight into key recovery methods, a critical aspect of managing large database systems, emphasizing the importance of maintaining data availability and optimizing transactions and data changes in real-time. Both commercial and research-related parallel systems are explored.