Comparison of immediate-update and workspace transactions: serializability and failure tolerance
S. Turc
Abstract
S. Turc
Abstract
A theoretical study of concurrency control and failure tolerance which includes both immediate-update (IU) and workspace (WS) transactions is presented. All previous formal approaches only consider IU transactions. A WS transaction first reads objects and updates them only in its private workspace; the objects are written only after the WS transaction commits. In order to examine execution correctness for both transaction models, it is necessary to reshape serializability theory. The framework constructed here handles both transaction and system failures and covers IU and WS transactions. The results show that the two transaction types impose different conditions on schedulers and recovery algorithms and deny the fact that WS transactions require optimistic schedulers and 'intention list' recovery. In comparing the two models, some histories of WS transactions which could not be obtained with IU transactions are given.>
A significance statement is not available in the OpenAlex record.
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.
A theoretical study of concurrency control and failure tolerance which includes both immediate-update (IU) and workspace (WS) transactions is presented. All previous formal approaches only consider IU transactions. A WS transaction first reads objects and updates them only in its private workspace; the objects are written only after the WS transaction commits. In order to examine execution correctness for both transaction models, it is necessary to reshape serializability theory. The framework constructed here handles both transaction and system failures and covers IU and WS transactions. The results show that the two transaction types impose different conditions on schedulers and recovery algorithms and deny the fact that WS transactions require optimistic schedulers and 'intention list' recovery. In comparing the two models, some histories of WS transactions which could not be obtained with IU transactions are given.>
Key concepts: Serializability, Computer science, Two-phase locking, Distributed transaction, Database transaction, Transaction processing, Workspace, Correctness