Active storage management for database systems
Chye Lin Chee, C. V. Ramamoorthy · University of California, Berkeley eBooks · 1995
The focus of this research effort is on studying research issues and prototyping of an active storage management system for use in database systems. The novel feature of the storage management system is that it has the ability to adapt to the data access characteristics of the application that uses it based on collection and analysis of runtime statistics. Adaptive reorganization is performed by the storage management system in a manner that optimizes the access patterns of the system for which it is used. Since the storage management system is intended for use in a transaction processing environment, primitive support for transactions and recovery is also provided; with more sophisticated features implemented by higher layers. Potential research contributions include dramatic performance benefits arising from read and write optimization, especially in a RAID environment. A log-structured storage system that naturally caters for write optimization is developed, along with a statistics collection mechanism to determine data access patterns of applications. Since the storage system is intended for use by database applications, primitive support for transactions and crash recovery is also built into the system. Read and write block-level locking on files are provided, as are primitive support for transactions and facilities for checkpointing and crash recovery. Versioning facilities are also provided across transactions, enabling users to retrieve committed data at any point in history. We explore various techniques to validate the effectiveness of our storage management system, including analytical modeling and prototyping. Analytical modeling is used in the early stages of our project to serve as a validation for our research, while the prototype is developed as proof of concept to verify the effectiveness of different rule mechanisms for our reorganization strategies. Performance results from our prototype show dramatic improvements in response time arising from prefetching and storage reorganization strategies in cases where access patterns are relatively predictable. Prefetching alone without storage reorganization speeds disk reads up to 33% over the basic write-optimized log-structured prototype in the best case. Storage reorganization without prefetching speeds disk reads up to 25%. The best results come from a combination of prefetching and storage reorganization, with performance gains of up to 40% in the best case.