X/Open Distributed Transaction Processing (DTP) Model

The X/Open Distributed Transaction Processing (DTP) model defines a specification for the management of transactions whose operations are distributed among different computers or among different databases from different vendors. DTP can be broken down into 3 core components:

• An Application Program (AP) for defining transaction boundires and specifies actions that constitute a transaction.
• Resource Managers (RMs, such as databases or Data Access Drivers ie ODBC, JDBC, ADO.Net for accessing them)provide access to shared resources.
• A Transaction Manager (TM) that assigns identifiers to a transactions, monitors the progress, and takes responsibility for transaction completion and fault recovery.

The TX standard defines the API used for communication between an application program and a transaction manager. The XA (eXtended Architecture) standard defines the two-phase commit protocol and the API used for communication between a transaction manager and a resource manager. A resource manager is a system service that manages durable data, e.g. a DBMS or file system. A transaction manager manages distributed transactions, which may span multiple resource managers. It manages two-phase commit, coordinating the decision to commit distributed transactions or roll them back, and coordinates failure recovery.

Whenever an application has a single function that needs to access or update the data in multiple transactional resources, it should use a distributed transaction. You can use a separate simple transaction on each of the resources, but this approach is error-prone. If the transaction in one resource commits successfully but another fails and must roll back, the first transaction can no longer be rolled back, so the state of the application becomes inconsistent. If one resource commits successfully but the system crashes before the other resource can commit successfully, the application again is inconsistent. An application moving data from one database to another requires a distributed transaction. Otherwise, the data may be duplicated (if the insert completes and the delete fails) or lost (if the delete completes and the insert fails).

Tuxedo

BEA Tuxedo provides the framework, or middleware, for building scalable multi-tier client/server applications in heterogeneous (dissimilar), distributed environments that extend from the Web to the Enterprise. Using BEA Tuxedo, users can develop, manage, and deploy distributed applications independently of the underlying hardware, operating system, network, and database environment.

In contrast to classic 2-tier client/server configuration of SQL servers, the Tuxedo brings 3-tier paradigm (clients, services, resource managers).

At the foundation of BEA Tuxedo ATMI is a proven, reliable transaction processor, also known as a transaction processing (TP) monitor. A transaction processor is an example of a 3-tier client/server architecture, where the transaction processor supports the application logic (represented by "services" between the GUI front-end and the back-end resource managers. Examples of resource managers are SQL databases, message queues, legacy applications, and other back-end services.

Tuxedo's Application-to-Transaction Monitor Interface consists of the Tuxedo transaction processor and infrastructure, and the ATMI API. The ATMI API is used to create a Tuxedo client and server.

A Tuxedo ATMI client collects a user request and forwards it to a server offering the requested service. The client "joins" a distributed Tuxedo application by calling the ATMI client initialization routine. The client can then exchange information with the server, sending and receiving data in typed buffers, and define transaction boundaries. The client "leaves" the application by issuing an ATMI termination function.

A Tuxedo server process oversees a set of services, calling them automatically for clients that request them. ATMI clients do not call servers by name; they call services. The server is not written as a complete program (i.e. with a standard main). Instead, the server's services are coded as a set of functions that are compiled with Tuxedo binaries to produce a server executable.

Distributed Transaction Coordinator (DTC)

The Microsoft COM and MTS interfaces were unified into a single run-time layer and renamed COM+. COM+ includes the transaction support that was created for MTS. Both include a system service, DTC, for coordinating distributed transactions and the two-phase commit protocol. Every machine running COM+ has a DTC system service.

For distributed transactions, each computer has a local transaction manager ie. the DTC on that machine. When a transaction does work at multiple computers, the transaction managers interact with other transaction managers via either a superior or subordinate relationship. These relationships are relevant only for a particular transaction. The root transaction manager, also referred to as the global commit coordinator, is the transaction manager on the system that initiates a transaction.

DTC introduces the notion of a resource dispenser which does two things. It manages a pool of connections to a resource manager (i.e. a DBMS in the context of Data Access) and it automatically enlists the resource manager in a component's transaction. The ODBC Driver Manager is a typical resource dispenser, as is an ADO .NET Data Provider.