The details of adapter architecture and design are explored in Chapter 4, "Adapter Reference Model." From a methodology perspective, the adapter architecture and design is no doubt impacted by the constraints identified in the analysis phase. The primary object of adapters is to encapsulate the application and expose integration functions and features. Hence, the adapter design should be driven by the right balance of long-term objectives enabling easier secure integration and short-term objectives of specific integration with specific applications and integration scenarios.
The complexities of interfacing with the application should not affect the long-term value of the adapter. Sometimes, this is hard to achieve with applications that have very closed architectures. Adapters for such applications can be very complex, and the benefits of integrating the application should be considered and weighed with the cost of developing complex adapters. If the application is strategic or mission-critical, then developing an adapter is always beneficial despite the possible complexities.
Adapter developers need to be some of the most creative programmers because they face challenging integration scenarios. Given the choice between constraining adapter functionality due to the underlying application architecture and developing complex integration adapters, the decision will be based on the significance of the application in the integration scenario. If the application is identified to be the master application that drives data synchronization actions, adapter developers may need to modify the application so it's more open to adapters.
Some of the design choices that adapter developers have to make involve the selection of point-to-point integration versus broker-enabled integration. Integration brokers are great tools for centralizing and managing complex integration scenarios involving intelligent routing of data and messages. However, you pay a price for the flexibility of using integration brokers. Sometimes when the integration scenario is simple, few applications are involved in the scenario (between three and five), and the volume of data is small, it might be better to develop a point-to-point adapter-based integration solution. This is more cost-effective and also easier to manage. However, the adapter design must be easy to migrate to a broker-enabled integration scenario without additional modifications in future.
Other design choices include persistence mechanisms that can include serialized objects, RDBMS, property files, and so on. Once again, the final decision should be guided by the foreseen integration scenarios. Obviously, large corporations need to plan for flexible integration platforms and adapter designs from the very beginning. Another design choice is the document model standard driving the data exchange mechanisms between adapters. Given the maturity and broad-based support for XML and XML-related technologies, it is logical to select XML as the document model. However, developers new to XML face a steep learning curve with many new technologies, and must assess the impact of learning on project estimates.
Selecting an Implementation Environment
The implementation environment for an adapter depends on various things, including the integration context (data integration, Web service, process automation) and existing infrastructure. The implementation environment for adapters usually comprises the operating system, servers, and database.
It is quite possible for the adapter to run on the same platform as the business application. Many times, adapters need to support remote execution. This is especially required when the two applications collaborating in the integration scenario are located in remote locations or on different hardware and operating platforms. The operating system(s) on which the adapters are expected to run also define the choice of supporting software-like databases and other servers. The selection of appropriate types and versions of operating systems is critical to the final adapter implementation environment.
For example, an adapter can integrate with a CICS application on the IBM OS/390 platform to use CICS transaction gateways on Windows NT environment to access the CICS application data. In this case, the adapter implementation is simpler because it involves only Windows NT instead of both Windows NT and OS/390. However, if the integration scenario needs the adapter to run closer to the CICS application, then the adapter needs to support remote execution on the OS/390 platform, so it is much more complex.
Depending on the type of adapter and the integration scenario, one or more servers may be required as adapter hosting environments or gateways. Application servers, database servers, Web servers, WAP servers, and transaction servers all become part of adapter implementation environments depending on the individual application architectures and the integration scenario.
Adapters need to store their configuration information in a fail-safe environment. Depending on the number of adapters and the frequency of changes to the configuration, a simple RDBMS-based configuration database or a high availability, data persistence solution may be required. Storing adapter configuration environments in a separate environment is a good design principle because it isolates adapter- generated control information (error messages, log files, and configuration data) from other application and system data.
Constraints Identified During Analysis
As developers begin to analyze the application architecture and design, it is quite common to identify serious constraints. Often, legacy applications (and sometimes even newer applications) are not designed with integration capabilities. Database stored procedures are not always thread safe or re-entrant. Many stored procedures do not identify the actual user invoking them. Database security and access are often defined by user requirements, and do not include application integration requirements. Allowing applications to access the database as a different type of users (invisible user) may require changes to the security policies used by the database administration teams.
Architecture constraints are the most difficult to solve during adapter development. If no APIs are available to the adapter developer, the design choices are significantly reduced unless the application is changed. Even then, refactoring application architecture and design is always prone to errors and introduction of new bugs. It is typical and wise to expect new bugs or resurfacing of old application bugs in the context of integration projects. It is important to remember that adapters cannot add new functionality to the application; they can only increase its integration capabilities. However, you can change or enhance the application functionality at the same time the adapter is being developed.
Adapter Hosting Environment
Adapters can be hosted in a multitude of environments, ranging from simple operating systems and application servers to sophisticated integration brokers. The difference between hosting environment and implementation environments are the supported services specific to adapters. For example, a J2EE-compliant application server can be an implementation environment; however, the JCA-compliant services define an adapter hosting environment. Adapters can be built with or without JCA support with the adapter architecture varying based on the adapter hosting environment.
Adapters can be hosted by different hosting environments. Some are sophisticated, like JCA-compliant application servers; and others are not so sophisticated, like the UNIX operating system. The level of sophistication expected is very specific to the requirements of managing adapters and providing adapters with system-level services. It is not always possible or even required to host adapters in an application server environment. Adapters can be hosted as UNIX processes, as standalone applications, as embedded components, or even as services and components of a distributed environment such as DCOM or CORBA.
Selecting the appropriate hosting environment depends on the integration scenario, requirements, and constraints identified. If the scenario is expected to achieve data synchronization between different databases, it is possible to host the adapters on the database servers, perhaps even including database triggers and procedures. The actual hosting environment can be selected based on the following criteria.
Level of Manageability
Adapters requiring a higher level of management services (dynamic configuration capabilities, graphical representations, dynamic load balancing, fail over capabilities, and so on) will require a sophisticated environment such as J2EE and JCA-compliant application servers or a CORBA-based distributed object environment. On the other hand, some adapters may not be complex or may be more static in their configurations. These could be hosted in a simpler environment, such as a Web server or a UNIX/NT-based workstation.
The volume of data, number of service messages, or number of workflow events to be processed can determine the hosting environment. Higher performance requirements will evidently require faster machines as well as high throughput environments. Higher-end application servers or those environments supporting clustering technologies will need to be considered.
Transaction and security requirements impact the adapter hosting choices on the operational side. Single sign-on requirements, distributed transactions, two-phase commit, and other transactional requirements must be provided by the hosting environment when needed. For example, when you're using MQSeries as an adapter hosting environment, you may need to develop some additional transaction and security services to meet the adapter requirements. MQSeries is one of the best messaging and queuing platforms, but does not perform the functions of a transaction monitor.
It is clear that adapter hosting environments need to be analyzed and selected based on a number of different factors. Some adapters can be hosted in a simpler environment, whereas others cannot function without a sophisticated adapter hosting environment. It is always a good design principle to develop scalable adapters capable of working in a multitude of hosting environments.
Building the Target Reference Model
In the previous sections, we have covered some of the important aspects of adapter analysis and design. Without a common reference model, the architecture of adapters probably will be different for each adapter. Consistency in design requires a common set of standards, and a reference model captures those critical parameters and design principles. Chapter 4 defines a target reference model for adapters. A similar model or a specialized version of this reference model should be accessible to adapter developers.
Reference models ensure a common design philosophy and a set of design patterns useful for developing adapters. The reference model is a good starting point when developing adapters. It provides you with a structural model of the adapter that can be extended and specialized for the specific adapter requirements without sacrificing a common infrastructure.