Performance considerations for an operating system transaction manager
Akhil Kumar, Michael Stonebraker
Abstract
Akhil Kumar, Michael Stonebraker
Abstract
Results of a previous comparison study (A. Kumar and M. Stonebraker, 1987) between a conventional transaction manager and an operating system (OS) transaction manager indicated that the OS transaction manager incurs a severe performance penalty and appears to be feasible only in special circumstances. Three approaches for enhancing the performance of an OS transaction manager are considered. The first strategy is to improve performance by reducing the cost of lock acquisition and by compressing the log. The second strategy explores the possibility of still further improvements from additional semantics to be built into an OS transaction system. The last strategy is to use a modified index structure that makes update operations less expensive to perform. The results show that the OS will have to implement essentially all of the specialized tactics for transaction management that are currently used by a database management system (DBMS) in order to match DBMS performance.>
OpenAlex reports 9 citations for this work. Citation counts describe recorded attention and do not establish research quality.
A contribution statement is not available in the OpenAlex record.
Method details are not available in the OpenAlex metadata.
Findings are not separately available in the OpenAlex metadata.
Limitations are not available in the OpenAlex metadata.
Application details are not available in the OpenAlex metadata.
Results of a previous comparison study (A. Kumar and M. Stonebraker, 1987) between a conventional transaction manager and an operating system (OS) transaction manager indicated that the OS transaction manager incurs a severe performance penalty and appears to be feasible only in special circumstances. Three approaches for enhancing the performance of an OS transaction manager are considered. The first strategy is to improve performance by reducing the cost of lock acquisition and by compressing the log. The second strategy explores the possibility of still further improvements from additional semantics to be built into an OS transaction system. The last strategy is to use a modified index structure that makes update operations less expensive to perform. The results show that the OS will have to implement essentially all of the specialized tactics for transaction management that are currently used by a database management system (DBMS) in order to match DBMS performance.>
Key concepts: Computer science, Transaction processing system, Database transaction, Transaction processing, Distributed transaction, Online transaction processing, Database, Transaction log