Service Contract Design Strategies

A key design principle for a Service Oriented Architecture (SOA) is the definition of a service contract. The contract clearly specifies the technical interface a client uses to invoke each service operation, as well as the expected behavior of the service. The contract is independent of the underlying implementation, so users of theservice don’t need to care about implementation details and are shielded from its changes over time. Implementers of the service benefit, too, as they have broad flexibility in determining the implementation strategy, and can fine-tune the implementation over time.

When designing data services, there are two main strategies for service design: contract-first service design and data source-first service design. Contract-first design is very much a top-down design strategy, while data source-first design is bottom-up. In contract-first service design, the technical interface is clearly defined first, using XML Schema to define the information flowing across the interface. The information objects defined by the XML schema, which are the primary subjects of these services, are often custom-designed, but sometimes they are drawn from industry standard schemas. If you work within an industry with a commonly used industry XML schema, chances are good that schema contains many definitions you can reuse for your own data services.

Data source-first design often fits the mental model of a data-oriented developer or architect who is intimately familiar with the physical data sources. In this scenario, the emphasis is more on exposing the data as is, combining data from multiple sources as necessary. The design process here follows a trail of building low level components, then selecting a set of them, and combining them together into a composite which is then exposed as a service. The final technical interface is a composite of many lower level components. The contract is simply derived as a result of the composition work. Some caution is warranted with this approach, as the resulting interface is essentially hard-wired into the back-end systems, making interfaces brittle in the face of change to data sources or client needs. The benefit to this approach is that it leads to extremely quick service implementations, and fits the mental model of many of those charged with implementing data services. Areas of development where many fine-grained services are required will benefit from these approaches.

Categories of Services

In previous blogs, I have discussed contract-first service design and the benefits to constructing services as information-rich, smart endpoints. In his most recent book, SOA: Principles of Service Design, Thomas Erl formalizes and extends these ideas by defining 3 categories of services based on the type of logic they encapsulate, the extent of potential reuse, and the degree of focus on a singular domain. The service categories are:

Entity Services – modeled around business entities like customer, purchase order, insurance policy, and invoice. Such a service typically includes the CRUD-like operations to Create, Read, Update, and Delete instances of the business entity. As I’ve mentioned in the past, entity services are ideal candidates for contract-first service design, with entities modeled in XML Schema, and that schema becoming the basis of the interface to create, read, update, or delete a business entity.

Task Services – services that generally model a business process in the enterprise. This type of service very often combines business logic and calls to other services, and because of its focus on a particular business domain, often has a low degree of reusability. Orchestration and choreography environments, along with environments that facilitate flow logic (XAware, for example) are common ways in which to create task services, although many are still writing code to stitch together services with business logic to expose “task” services. Task services also typically involve operations on more than one entity. When processing a purchase order, for example, a process may require validating a customer (using the “customer” entity service), generation of a unique purchase order number (using a order number generator service – a utility service), and adding the order to the provisioning system (using the purchase order entity service). A requirement to access multiple entities is an indicator that your service is a task service. Task services are often referred to by other names, among them: business process services, process services, task-centric business services, or orchestration services (when using orchestration tools).

Utility Services – services that are technology-oriented rather than business focused. Utility services typically provide common and reusable functions that cut across functional domains, such as event logging, notification, and unique number generation. Utility services are often referred to as “technology services” or “infrastructure services”.

The service category definitions are just the beginning in Erl’s book. The heart of the book discusses a palette of principles that guide a designer through the complex activity of creating services with appropriate characteristics for a particular use case and environment within an SOA. This is a book that I highly recommend, and I plan on referencing Erl’s ideas frequently in future blogs.

The Long Tail of Services

The “Long Tail” moniker has been used (some would say overused) in marketing and mass media over the last couple years. Now, as different use cases for SOA evolve, the metaphor seems appropriate to explain the growing use of services outside of the core domain of SOA, business process implementation and integration. If you’re not familiar with the term “long tail”, it is used to describe how companies like Amazon, iTunes, and Netflix have monetized the huge number of low-volume products they carry in inventory. While traditional retailers make most of their money selling only the blockbusters, these companies have discovered how to make significant money from the products that sell in very small quantities. In “The Long Tail” diagram below, the green area (the head) represents the “blockbusters”, a small number of high volumes sales items. The yellow area is the tail, representing a large number of items selling at very low volume. The shaded area under the curve can be thought of as revenue. You can imagine that if the tail is long enough, its revenue is significant, and can even exceed that of the head. Chris Anderson coined the term “The Long Tail”, and wrote an excellent book on the subject, which you can read more about here.

The Long Tail

The Long Tail of Services

The long tale metaphor also aptly explains and helps justify the phenomenon of minimalist services created for singular purposes. While the SOA mantra is to create services for reuse, the rise of Web 2.0 technologies like AJAX and Flex has created a different niche for the development of services. I’ve ranted in the past that this style of service is not very service oriented.

The Long Tail of Services: Single-use and low reuse services may outnumber those designed for reuse.

In “The Long Tail of Services” diagram, the horizontal axis represents individual services, and the vertical axis shows “instances of use”. The head area represents services that are reused many times. The tail area shows services with ever decreasing reuse, culminating with services used only a single time (single-use services).

Services for Web 2.0 Applications

Building services for consumption in a user-facing application is quite different than building services to feed core business processes. If you download and use any of the AJAX-like environments, or Adobe’s Flex, and build a sample display fed from XML data, you’ll likely notice the XML structure tends to follow a simple, almost table-like structure. This is very simplistic compared to the information-rich XML structure you might expect in real-world business objects like a purchase order, invoice, or insurance policy. The latter are the domain of SOA proper, the subjects manipulated by services while moving through business processes.

At first glance, I would say the Web 2.0 style services are high-quantity, commoditized, point services fulfilling a specific, singular purpose. Don’t expect reuse from these, as they were not designed with reuse as a design criteria. But thinking further about this, I now feel that a small service like this may actually be a good candidate for reuse. The reason: its visualization is the main manifestation of its existence. In fact, a user who sees the visualization of the information is probably going to be the one to recognize that it can be used in other situations. Isn’t this a much more powerful reuse driver than a UDDI query, or worse, a WSDL file sitting on a web server or file system somewhere? Even with a nice WSDL viewer, it takes imagination to envision the data reused in another application. If this visualization aspect does turn out to be a big factor in reuse, perhaps that will drive a movement to publish services with default visualization components. It sure makes sense to be able to visualize a data structure in a display composition, rather than simply viewing an XML structure.

Other Aspects of The Long Tail of Services

The Long Tail of Services diagram is also a good context to help rationalize other aspects of service oriented design. I can envision where on the graph you would expect to apply the most governance energy (services whose intent is high reuse), and where you might expect to use light-weight REST or POX rather than SOAP as a delivery mechanism (in the long tail). Debates such as REST versus WS-* are still simmering, but The Long Tail of Services may at least provide a different dimension on the discussion, providing a context where one point of view makes more sense relative to one section of the curve.

Credits: The Long Tail picture was created by by Hay Kranen / PD, and annotated by me.

Web 2.0 Powered by SOA

Mark Scrimshire blogged about attending a Ray Valdes presentation on Portals and Mashups (see the blog here). The article nicely explains how Web 2.0 and SOA technologies combine to produce a new generation of portals, fed by information culled from throughout the enterprise. The separation of concerns between Web 2.0 and SOA technologies can be summarized in a single sentence:

“Information is delivered through a Portal, Not by a Portal.”

Simply put, Web 2.0 provides the visual set of user-customizable portlets, while SOA supplies the information for the visual screens. The information is typically provided in a service-oriented manner, with a lightweight REST approach gaining preference over SOAP for the service delivery mechanism. Interestingly, over 50% of the audience was specifically using an SOA strategy, but only about 10% felt the process was successful. Deployment cylces were reportedly very long, on the order of 2 years before demonstrating value.

The project I lead at XAware.org fits nicely into this Web 2.0 and SOA approach. XAware lets you quickly build and manage data services from any number of enterprise data sources, then deliver those services using a lightweight REST or POX (plain old XML over HTTP) transport. Enterprise data is complex, with the typical company housing hundreds of data sources. The complexity of this environment explains the long development cycles mentioned above. XAware specifically targets the complexity by mapping different sections of your XML-based service contract to different back end systems, using the Eclipse-based design environment. XAware users seem to agree, as the user base is growing rapidly, especially where many data sources and complex information structures are involved.

Eclipse Conned By Microsoft

I attended EclipseCon 2008 this week in Santa Clara, CA. Wednesday's keynote speaker was Sam Ramji from Microsoft, talking about Microsoft's growing participation in Open Source initiatives. He began by confirming rumors that a new initiative, "Supernova", would have Microsoft purchasing the Eclipse organization, and acquiring all intellectual property of Eclipse. A couple weeks shy of April Fools Day, Sam still got lots of laughs on that one, from an understandably skeptical crowd. He talked about his group, the Open Source Software Lab, whose broad purpose it is to assist the open source community with interoperability with Microsoft software. Microsoft recognizes that more and more open source software is running on Microsoft platforms, and they want to continue improving the platform to facilitate this growth. Microsoft sponsors a .NET open source forge at www.codeplex.com. Sam's open source work is closely aligned with Microsoft's recently announced interoperability principles, which signals a more open direction for Office, SQL Server, SharePoint, Vista, and .Net.

While I believe Microsoft is in a different galaxy when gaging their level of Open Source commitment, I do believe they are moving in the right direction. They are doing just about what you might expect from the largest, most successful commercial software company in the world. They are encouraging interoperability to increase their own platform market. But Sam asked for patience from the EclipseCon crowd, noting that in 1995, everybody assumed Microsoft missed the boat on the internet. Ten years later, they are one of the dominant players in that space. Sam propositioned that with Open Source too, Microsoft may be starting slow, but ten years from now, Microsoft may be known as a dominant Open Source player. I am skeptical on this point, as I believe were most of the attendees. But Sam's interoperability initiative is at least a good step in the right direction.

I attended a number of EclipseCon sessions Wednesday morning, covering Service Component Architecture (SCA) tooling, JBoss's jBPM, and Eclipse metadata management initiatives. The SCA tooling is part of the new SOA Tools Platform, one that is emerging as an important member of the Open Source SOA ecosystem. XAware plays well with all these tools, interoperating at a number of levels, ultimately providing information-rich services as the foundation of an SOA initiative.

As if a Microsoft keynote speaker at an Open Source event weren't unusual enough, things grew really strange when I had a brief encounter with a flamboyant Marilyn Monroe while exiting the convention center. A number of her bouncing body parts looked totally fake, not least of which was her platinum blonde hair, obviously a wig. It turns out she was stumping for NetBeans among the Eclipse faithful. For those who don't know, NetBeans is the Sun-sponsored competitor to Eclipse, also an Open Source, pluggable Interactive Development Environment (IDE). Marilyn smiled broadly and handed me a disk, then invited me to the NetBeans party at the Techmart building next door. Instead of heading to my car, I took a left turn towards the Techmart (guess I'm a sucker for such advertising gimmicks), and took a quick walk through. The NetBeans crew had free beer and other refreshments along with demonstrations of the NetBeans environment. I didn't have much time to hang around, but I appreciated the amusing marketing ploy, a perfectly targeted campaign aimed at EclipseCon attendees.

Urban Sprawl in SOA Land

Service development is about to explode in the enterprise, but these are not the custom-crafted, high value, reusable services you might see cruising the pristine avenue of your company’s SOA center of excellence. No, this will be massive. Instead of nicely architected brownstones with clean lines and aesthetic appeal, you’ll brush with small, hastily constructed, one-off services as far as the eye can see. Its urban sprawl in SOA land, and things are about to spiral out of control.

Where will this “urban sprawl” of services come from? The new killer app for SOA is Rich Internet Applications (RIA). Vendors and pundits are vocal about the perfect fit of RIA and SOA. Environments like Flex , Appcelerator, and many of the AJAX-based development environments like Tibco GI seem to perfectly match RIA and services.

It seems like an excellent fit on the surface, using services to populate sections of the display and to perform major operations like "process order". But management of the services is a glaring risk area. SOA generally strives to create coarse-grained services aligned with steps in a business process. You can visualize each service as an activity node on a flow graph. From this basic idea you get many of SOA's benefits: coarse grained, loosely coupled, reusable services that can be re-orchestrated into new business processes.

In the RIA-SOA scenario, left unchecked, I envision a high number of very granular services. Most of these will be tightly coupled, not very reusable, and RPC style. Maybe there is nothing wrong with this, but it is a significant divergence from the core principles of SOA.

I personally like the idea of RIA-SOA, but the trick will be to encourage and enforce SOA principles in the development of this new style of application. Otherwise, expect rampant proliferation of low-value services.

I can envision certain RIA development styles that are more “service oriented” than others. A quick example can illustrate this. Suppose an auto insurance company needs a new policy administration system. From the user perspective, an agent may enter a policy number and expect to see the auto policy in a multiple-tab view. Lets say that one tab shows the policy owner (name, address, contact info, etc), another tab shows coverage information (dollar limits for damage, medical, liability, etc.), and a third tab shows premium information.

The designer of such an application has many alternatives to supply services to power the screens. The primary characteristic differentiating the approaches is how coarse-grained the services are. For illustrative purposes, lets talk about three degrees of coarse-grain-ness: coarse-grained, medium-grained, and fine-grained. A coarse-grained approach would provide services to “get policy” and “update policy”. The RIA application would call “get policy” when the agent enters the policy number. Each tab would read and update appropriate sections of the single, shared policy structure. The agent would click an update button to invoke the “update policy” service under the covers, which sends the policy structure to a service which performs the needed updates. This strategy not only limits round-trip calls to the server, but conforms to SOA norms by encouraging development of more coarse-grained services that are likely more reusable and more loosely coupled.

A medium-grained approach would provide us with a separate service invocation for each of the tabbed panels. Three tabs leads to 6 services to read and update appropriate portions of the policy. The application would likely exhibit better perceived performance, as the user could view the first screen as soon as the first service (get policy owner) completes. While we have potentially 6 round trips instead of 2 in the coarse-grained case, the update operation may deal with less data, since we only have to update individual portions that changed. These services seem reusable and loosely coupled as well, as they are grouped around business information that likely is used by many applications.

A fine-grained approach leads to a larger number of services. Imagine you design your GUI top-down. You go along building screens and stubbing out services you will later create to populate the fields and react to user events. Your services are precisely defined to fit the specific needs of each screen. It a bit harder to imagine this style with our insurance example, but perhaps we define a service to retrieve name, another for address information, and another for contact information. Each section of each screen gets its own service. We have many more services now, but we increase the opportunity for perceived visual refresh performance. Another benefit is that the service definitions fall out immediately from the information sections of each screen, making the design effort straightforward. But the sheer number of services will cost us in terms of management overhead. And these services are so closely aligned to screen layout as to make reuse outside the RIA environment very unlikely.

From this example, we see that as we move down the scale towards more fine-grained services, we increase the opportunity for perceived visual refresh performance while we increase the number of round-trip service invocations to the server. We also increase the management overhead by introducing dramatically more design artifacts.

While the policy administration example shows tradeoffs in the service design approaches, in the context of service orientation, the characteristics are not quite so arbitrary. To be service oriented, you must strive to provide coarse-grained services that are reusable and loosely coupled. To a large extent, if you think in terms of defining the information payloads for services at a higher level, from the business analyst perspective, you will achieve better reuse and loose coupling. On the other hand, if your service interface is defined by a specific set of screen fields, you have tightly coupled your service to its consumer. When the screen changes, your service interface must change. Such a service has a singular purpose, will not be reusable in any other context, and so by definition, is not service oriented.

The message to RIA designers and architects is simple. Yes, industry needs you to combine RIA and SOA to achieve a new level of productivity and richness of user experience. But you should take up of the cause of SOA while you build your RIA. Be service oriented so your services are loosely coupled. Be service oriented so your services can be reused elsewhere. Your efforts will be a productivity multiplier within your organization, bearing fruit from your work long after your current project. Be service oriented, and in the process, you can stop the sprawl.

4 Generations of Services

While working at MCI in 1995, I was exposed for the first time to the benefits of service orientation. I was responsible for the order entry application for 1-800-MUSIC-NOW (which turned out to be a marketing flop), in which call center agents needed to process credit card payments for music CDs. Fortunately for me, MCI was already accepting credit card payment for long distance service. Even more fortunate was that a forward-thinking architect working in the credit-card processing group in Nashville had anticipated future uses of the software his group had developed. What we had feared would require several months of effort turned into about a week of coding to properly access the TCP/IP-based credit card authorization service.

I am sure my experience at MCI is not unique. Across industries, smart architects have been exposing functionality as reusable services. This got me thinking about the how service oriented architectures have been evolving over the years. Those of us working to expand capabilities in SOA tooling have recognized that services and SOA have been around far longer than the "web service standards" that have made SOA so popular over the last few years. It is apparent that in the software industry, we have gone through phases of maturity in our service-oriented implementations.

The long term vision of SOA is to enable rapid assembly of applications by orchestrating services into new business processes. Unlike the modest cost savings I experienced at MCI, the cost reductions achieved in a mature SOA could be immense. The traditional 12 or 18 month development cycles involving armies of analysts and developers are replaced by a business analyst drawing a flow graph of the new process, then clicking a button for deployment to enterprise-ready infrastructure. The "service oriented" approach is measured in days or weeks using a fraction of the engineering resources.

The industry is a long way from realizing the SOA vision, of course, but service orientation has certainly evolved a good deal in the past decade. To better understand this evolution, it is helpful to categorize technologies into generations. For example, programming languages have been categorized as they matured, from machine code to assembly, then FORTRAN, C/C++ and Java, and finally 4GLs like PowerBuilder, SQL, ColdFusion. Each generation achieved higher levels of abstraction and more statement power, making it ever-easier to translate process descriptions and algorithms into machine-executable form.

As service orientation has matured, a number of generations of services have emerged. These generations are defined loosely by how well they support or implement the major characteristics of SOA: standards-based, loosely coupled, coarse-grained, and business-orientation enabling an analyst to understand and manipulate it. Early generations reflect less of these core SOA values, and later generations are more comprehensively based on these values. Looking at the landscape of services over the last 5 years, I see 4 generations of services:

1st Generation Services - Simple services coded in a 3GL language like C, C++, C#, or Java, which don't use modern service standards like WS-* or REST. These services tend to tightly couple the consumer with underlying resources. Older distributed computing technologies like CORBA and DCOM fall into this category as well.

2nd Generation Services - Services that are standards-based and fairly simple, like implementing an operation to retrieve, modify, create, or delete a data set on a database. These services can often be auto-generated from other sources, such as from a Java or C# class, an EJB, or a database query. These services tend to reflect a method on an object, or expose an underlying implementation strategy like a relational table. They are easy to create, but because they are technology-oriented rather than business-oriented, they are unwieldy to use directly in a business process. Instead, they require combining with other services and logic to provide the proper level of granularity for orchestration.

3rd Generation Services - Truly "service-oriented", these services are aligned with a step in the business process. Loose coupling is achieved by explicitly defining data formats that are the payload of the service request and response, and these formats are driven by analysts that know the business process at hand, not by technologists attempting to optimize execution times and storage requirements. These services are often created by stitching together and transforming 1st and 2nd generation services to achieve a coarse-grained service that is suitable for orchestration and at the same time achieves loose coupling.

4th Generation Services - These are 3rd generation services that are institutionalized as managed, secure, governed, and reusable services. 4th generation services involves an ecosystem of SOA-aware technologies and procedures, which allow construction and management of business processes and higher level services. Given time to achieve 4th generation services, a company will maximize the benefits of SOA, enabling them create and modify business processes quickly to meet the changing demands of the business.

The precise boundaries of the generations and the metrics used to categorize a service can certainly be debated. But the concept is valuable in that it helps clarify the goal, to ultimately maximize the benefits of an SOA. These concepts also help justify implementing services in projects that have more modest goals. Maybe you know a function will be re-used across departments. Or, you anticipate that your basic service will be used by a 3rd generation service, thereby growing the foundation that will become an SOA based on 4th generation services. We don't need it all today, but we can derive concrete benefits in the short-term while maneuvering towards an even more attractive set of benefits in the future.