Semantics-based recovery in transaction management systems
Eliezer Levy
Abstract
Eliezer Levy
Abstract
A cornerstone of the transaction paradigm is the notion of atomicity. The principle that forms the basis for obtaining transaction atomicity in most contemporary database systems is to prohibit transactions from accessing uncommitted data. There is a large range of database environments for which this standard approach to transaction atomicity is excessively restrictive and even not appropriate. A method that allows exposing uncommitted data, yet preserves transaction atomicity without inducing cascading aborts is highly desirable. Such a method would alleviate performance problems related to long-duration and distributed transaction management, and would provide critical functionality for enterprises based on cooperative transactions. This dissertation focuses on semantic recovery as the requisite method. Semantic undoing, referred to as compensation, is carried out by a compensating transaction which is associated with a specific forward transaction. A compensating transaction faces the intricate task of undoing its forward transaction while obliterating the effects of others transactions to a minimal extent and preserving data consistency. Only with the aid of the specific semantics of the application at hand can this task be accomplished. Compensation, and semantic recovery in general, can be utilized in the realm of distributed transaction management systems. Supporting atomicity of multi-site transactions in a distributed system is equated with long-duration delays, blocking, and loss of the local autonomy of the individual sites. The two-phase commit protocol embodies these deficiencies. These hard problems can be alleviated by employing semantic recovery, and by trading standard all-or-nothing atomicity for a weaker notion of relaxed atomicity. Facing the relevant impossibility results in distributed computing, this new direction is well justified. Relaxed atomicity is characterized by an asynchronous process of recovery from decentralized and uncoordinated local decisions as to whether to commit or abort a multi-site transaction. This recovery process finally leads to a unanimous outcome. Relaxing standard atomicity interacts in a subtle way with correctness and concurrency control issues. Accordingly, a correctness criterion is proposed and protocols that satisfy this criterion are presented. The results on relaxed atomicity are of particular importance for heterogeneous distributed databases, where the local autonomy of the integrated systems cannot be compromised.
OpenAlex reports 3 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.
A cornerstone of the transaction paradigm is the notion of atomicity. The principle that forms the basis for obtaining transaction atomicity in most contemporary database systems is to prohibit transactions from accessing uncommitted data. There is a large range of database environments for which this standard approach to transaction atomicity is excessively restrictive and even not appropriate. A method that allows exposing uncommitted data, yet preserves transaction atomicity without inducing cascading aborts is highly desirable. Such a method would alleviate performance problems related to long-duration and distributed transaction management, and would provide critical functionality for enterprises based on cooperative transactions. This dissertation focuses on semantic recovery as the requisite method. Semantic undoing, referred to as compensation, is carried out by a compensating transaction which is associated with a specific forward transaction. A compensating transaction faces the intricate task of undoing its forward transaction while obliterating the effects of others transactions to a minimal extent and preserving data consistency. Only with the aid of the specific semantics of the application at hand can this task be accomplished. Compensation, and semantic recovery in general, can be utilized in the realm of distributed transaction management systems. Supporting atomicity of multi-site transactions in a distributed system is equated with long-duration delays, blocking, and loss of the local autonomy of the individual sites. The two-phase commit protocol embodies these deficiencies. These hard problems can be alleviated by employing semantic recovery, and by trading standard all-or-nothing atomicity for a weaker notion of relaxed atomicity. Facing the relevant impossibility results in distributed computing, this new direction is well justified. Relaxed atomicity is characterized by an asynchronous process of recovery from decentralized and uncoordinated local decisions as to whether to commit or abort a multi-site transaction. This recovery process finally leads to a unanimous outcome. Relaxing standard atomicity interacts in a subtle way with correctness and concurrency control issues. Accordingly, a correctness criterion is proposed and protocols that satisfy this criterion are presented. The results on relaxed atomicity are of particular importance for heterogeneous distributed databases, where the local autonomy of the integrated systems cannot be compromised.
Key concepts: Atomicity, Distributed transaction, Computer science, Database transaction, Compensating transaction, Two-phase commit protocol, Transaction processing, Distributed computing