Common message format

Common messages refer to messages that are transferred across the tiers of a n-tier architecture.  Typically, user-interface information is captured, sent through to a controller tier, processed in the business or application layers and then passed on to a persistence layer or back-end legacy system.  During each of these transfers, a message is being exchanged between tiers where each may have a different format.  The key thing is to agree on one standard for a common message format within the enterprise so as to overcome any format translation overhead where an Enterprise Service Bus (ESB) is not used or when its use is deemed expensive in terms of format translation.  The use of an ESB will take care of many of these mediations, transformations and routing.  This is the Integration layer as depicted in the SOA layers model.

In some cases, it may suffice to agree on the outgoing and incoming message formats.  The issue of a common message format is a key architectural decision: you may choose to "roll your own" as specified here, adopt industry models such as travel industry's OTA XML or adopt non-industry specific models such as those defined by OAGIS.  In some cases, the decision will be to use a common enterprise message format that updates fields in the message and passes it on to the next tier for further processing.  If such a common scheme cannot be achieved due to political factors, then adapters that translate messages to an internal common message format can be designed.  They can also be leveraged as part of an ESB.

Common Message format and data architecture

In general, services should not indicate anything about the underlying data models.  Rather, they should be used to encapsulate the underlying data models whose data stores are leveraged by the service components that realize the services.  Thus, services that are view, edit, delete, add search and other operations on a database may not be good candidates for services, but could be used as underlying component operations as they are used today.

Existing data architectures that define conceptual, logical or physical data models are required sources in defining common message formats.  The definitions of common message formats should be coordinated with data architecture efforts and data models.  This analysis will ensure the availability of proper data stores and schemas for the service components that will realize new services.  Existing data architecture will be enhanced to accommodate new services if new data needs to be added to the underlying systems.

In many enterprises, existing systems often reflect the existence of silos and data islands that collaborate through batch processes.  Migrating away from isolated data stores is possible through the means of services.  Identifying a service provider's data sources is accomplished during service realization.

ISV Considerations: Logical Data Model needs to accommodate the pre-defined, often implicit data models embodied in ISV packages.  Thus message transfer between packaged applications and the existing data models need to occur.  This is often done through APIs offered by the ISVs. In a SOA, adapters to these ISV data models become important, especially if the ISV is not exposing its underlying data and functionality through services.

Note that in some cases where the ISV data model is accessible, it may be possible to customize the model to accommodate the messages required to support the identified services.  Conversely, if the data model is not accessible, service messaging may be constrained by the data model contained within the ISV.  A mediation mechanism can also be employed to curtail the problem.  Mediation, such as that provided by an ESB can be used in this context to support interfacing with ISV packages.  The ISV data model may also dictate additional attributes that are required above and beyond those required to implement the service.

Common message format with all relevant services

In the concept Concept: Domain Design, the notion of domain modeling was outlined, similar to the notion of an analysis model or business-analysis model in representing core concepts from the business domain in a technology independent manner. It is clear that the messages used by services are technology aware (if not technology-specific in the case of XML Schema used for Web Services) in the same way the database schema used to store the domain data is technology specific within the service. In fact, we may consider the following relationship.

This demonstrates the relationship between the domain model used for discovery of the key domain elements and the message model as the realization of the domain model as a set of elements passed into and returned from services.

The following is a typical Java/component model where we can see the separation of the Interface from the Class and the inclusion of "accessor" functions to get and set the value of the state variables. This is a very common approach, but it has the disadvantage that, if the consumer and component are in different address spaces or on different machine, the cost of communicating each call is high in terms of accessing the entire state of any one component.

Another issue is the relationships between components, the notion that an Account has a set of Customers is difficult to develop in this style and usually ends up in managing lists of identifiers used to retrieve individual objects.

In developing a service model we might use a data-driven service identification approach leading to the specification of an AccountMgr service and a MeetingMgr service. The first service specification acts as the central location for managing all accounts and contacts. In fact, the core data model for the Customer Relationship Management (CRM) solutions was built using this and other services. The second service has been separated because it can be used by the CRM solutions and other solutions for booking meetings and it will interface with the enterprise Groupware applications.

The following is a sample from the model; it shows the service specifications, the messages can be assumed from the domain model above.

When thinking about messages, there is a natural tendency to consider them to be simply the parameters to operations. This is made more likely because the UML representation of services uses operations with parameters and the Web Services Description Language (WSDL 1.1) uses a similar approach. However, when thinking in terms of Services and Service Specifications, it is more helpful to think in terms of messages as reusable elements that are either produced by or used/consumed by a service operation. In services parlance, the operation simply becomes a message exchange, albeit a named exchange on a service that is distinguishable from another exchange that may use the same input and output messages.

The notion of message exchange patterns has been of interest in the Web services standards world, as a part of the analysis of the use of services in developing standards to support their specification. A message exchange pattern names a particular combination of produced, used, or consumed message between two services (or between a service and a consumer) and provides a common vocabulary for service designers to describe operations on service specifications.

The following lists common exchange patterns that can be used in the definition of service specifications. Such patterns are usually found during the modeling of Service Collaborations.

Synchronous Request/Response: This is in effect a traditional method invocation where the service consumer sends a message to a service and then blocks waiting until a reply is received from the service.

One Way Message: In this case, the consumer simply sends a message to the service, not waiting for or expecting a reply. This pattern can be considered an asynchronous method call with no reply type, meaning that the service consumer continues execution after the message is sent, rather than waiting for the service to process the message.

Notification: In this case, the service is responsible for sending messages back to the consumer (usually another service). To accomplish thi,s the consumer has to have somehow registered with the service so that the service knows where to send the notification messages.

Asynchronous Request/Response: This is a combination of the one-way message and notification. The service consumer sends a message, including a reply-to address. When the service completes its processing, it calls the originator back. The fact that service consumers send the first message in an asynchronous manner does require them to keep track of all sent requests so that responses, when received from the service, can be correlated to the original request.

Publish/Subscribe: Again this is a combination. A service consumer registers interest in a "topic" with a publication service. Other services or service consumers publish messages (send messages) to the publication service identifying the topic associated with the message. If the topic matches previously registered consumers, they are notified of the new message. In this case it is possible to very loosely couple the services participating. Any consumer or publisher only needs to know the location of the publication service and new consumers can be added to the solution without significant effort.

Services are intended to provide large-granularity operations. As such, the messages that flow in and out of such operations tend to be large-grained also. This concern was originally highlighted early in the deployment of Web Service solutions where the use of HTTP as a transport, SOAP as a protocol, and XML as a wire format tended to lead to relatively slow responses and very high bandwidth requirements. For example, consider a request for a stock quote from a service. A simple stock quote was often demonstrated in the early Web Services days. The ticker symbol is four characters and the response is a decimal number. In an RPC style, binary protocol we might expect that the message identifier might add some overhead, let's say 8 bytes, and so we could expect somewhere in the region of 8+4 for the request and 8+8 (for a high precision decimal) in response. With HTTP/SOAP, we might expect something of the following form:

  SOAPAction: "http://www.webservicex.net/Quote"


  User-Agent: MyAgent 1.0


  Content-Type: text/xml; charset=UTF-8


  <?xml version="1.0" encoding="UTF-8"?>


  <soap:Envelope


   xmlns:soap="http://schemas.xmlsoap.org/soap/envelope/"


   xmlns:tns="http://www.webservicex.net/">


   xmlns:xs="http://www.w3.org/2001/XMLSchema">


   <soap:Body>


    <tns:Quote>


     <tns:Symbol>IBM</tns:Symbol>


    </tns:Quote>


   </soap:Body>


  </soap:Envelope>

  HTTP/1.1 200 OK


  X-Powered-By: ASP.NET


  Connection: close


  Content-Length: 522


  X-AspNet-Version: 1.1.4322


  Date: Mon, 21 Mar 2005 00:34:21 GMT


  Content-Type : text/xml; charset=utf-8


  Server : Microsoft-IIS/6.0Cache-Control: private, max-age=0


  <?xml version="1.0" encoding="utf-8"?>


  <soap:Envelope


   xmlns:soap="http://schemas.xmlsoap.org/soap/envelope/">


   xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance">


   xmlns:xsd="http://www.w3.org/2001/XMLSchema">


   <soap:Body>


    <QuoteResponse xmlns="http://www.webservicex.net/">


      <Quote><Last>89.28</Last>


      </Quote>


    </QuoteResponse>


   </soap:Body>


  </soap:Envelope>

Early adopters of Web Service Technology came to two conclusions. First, services were optimized for a small number of operations providing data as documents rather than the more complex style of traditional component models. This has the advantage of amortizing the overhead of the protocols across a bigger actual data payload. Also, between services in an enterprise, at least between services within the same solution, smaller,simpler protocol bindings were chosen and HTTP/SOAP was reserved for situations where it was necessary, such as interfacing to services outside the enterprise.

This notion is not entirely new. Even in the component world, the Value Object pattern, or the J2EE Service Facade are both approaches to reduce the number of communications round-trips between client and server. Both use the notion of sending a complete copy of the component state to the client rather than using the traditional accessor functions. For services, we may also like to consider the fact that services are being developed that are more closely aligned with business models, especially business process models. As such, messages come to reflect common business documents in the same way EDI Transaction Sets (Electronic Data Interchange) represent business documents such as orders, invoices, shipping notices, and so on.

In general, the use of large messages is valuable in overcoming communications performance, although in some cases the large data message can be a problem. For example, in the SOAP messages above, we saw how the message size, using HTTP, SOAP, and XML significantly increased the size of the data. This has been a complaint of early systems built using Web Services technologies. On the other hand, these issues have allowed us to learn some interesting lessons such as considering performance in terms of code performance, message design, and protocol choice as an early design activity.

One important aspect to consider is that as we are moving large chunks of state from service to consumer or service to service, these messages actually represent stale snapshots of the service state. So, one of the considerations is to explicitly manage this "staleness" by identifying the time the data can be considered reliable or to "lease" it to the consumer such that it expires after some amount of time. For more information, see the white paper Using Service-Oriented Architecture and Component-Based Development to Build Web Service Applications.

Another topic that has to be considered is the caching of content. Caching is usually a concern that is dealt with as a performance optimization for applications, but in an service-oriented solution, the distributed nature and message-based communication lends itself well to the insertion of caches between consumers and services. These caches are not the typical database caches used for optimizing queries, but more like caches used in web servers and web proxies. In fact, in the case of Web Services using HTTP and SOAP, these proxies may be used as caches to serve up service responses in certain situations.

The issues however, in regard to the use of any cache, are really how the cache understands the policies used to serve content from the cache and how a service can invalidate the cache. The technical infrastructure used to host and manage deployed services should provide caching capabilities. One area of service policy we expect to see in the future is the provision of cache management information.