Building SOA your way

Every enterprise needs to find its own balance between complete, scalable architecture and simply building a service-oriented architecture that works

A fault line runs beneath the groundswell that began a few years ago with XML Web services and continues today as SOA (service-oriented architecture). True, nearly everyone agrees that XML messaging is the right way to implement low-level, platform-agnostic services that can be composed into higher-level services that support enterprises business functions. Yet, here’s also a sense that the standards process has run amok.

IBM, Microsoft, and others have proposed so many Web services standards that a new collective noun had to be invented: WS-* (pronounced “WS star” or sometimes “WS splat”). The asterisk is a wild card that can stand for Addressing, Eventing, Policy, Routing, Reliability, ReliableMessaging, SecureConversation, Security, Transactions, Trust, and a frighteningly long list of other terms. Surveying this landscape, XML co-creator Tim Bray pronounced the WS-* stack “bloated, opaque, and insanely complex.”

It wasn’t always so. Simple forms of XML messaging were succeeding in the field long before any of these standards emerged. At InfoWorld’s SOA Executive Forum in May, Metratech CTO Jim Culbert described how his company’s service-oriented billing system worked back in the late 1990s. The messages exchanged among partners were modeled in XML and transported using HTTP with SSL encryption -- the method still used for most secure Web services communication today. Seybold analyst Brenda Michelson, who was then chief architect at L.L. Bean, tells a similar story about that company’s early experience with Web services.

Two factors were prominent at the time. First, the Web offered a simple, pervasive integration framework, one later promoted to the status of architecture and assigned the label REST (Representational State Transfer). Second, XML provided a universal way to define services in terms of the data they produced or consumed, rather than in terms of the code that produced or consumed the data. In combination, these factors were -- and still are -- powerful enablers.

Cranking Up Complexity

How, then, did we arrive at WS-*, which Culbert and others say is a cart that's gotten way ahead of its horse? One theory holds that the heavy-hitting vendors, working closely with key customers and partners, have ratcheted complexity up to a level that only they will be able to sustain. Because those specs are so far ahead of what most users need today, their development hasn’t been an organic process driven by well-known requirements.

Patrick Gannon, president and CEO of OASIS, the standards body now coordinating a number of the WS-* specifications, reluctantly agrees that users should have been more engaged from the beginning. “I wasn’t involved in creating those specs without formal user requirements on the table,” he says. “But I’m a pragmatist; the specs are there.”

Another view holds that industry heavyweights, who have paid their dues when it comes to security, transactions, and reliable messaging, are indeed qualified to translate their experience in these matters into the language of XML. TN Subramaniam, director of technology at RouteOne, which makes software that streamlines credit management applications on behalf of car dealers, learned that lesson the hard way. At one point he began drafting his own spec for single sign-on, only to abandon it when he discovered SAML, which his joint-venture partners enthusiastically adopted because all their identity management vendors -- including Netegrity and Oblix -- were supporting it.

"What are the chances,” Subramaniam asks, “that five architects meeting every other day will iron out all the possibilities, versus having a committee thinking it all through in great detail with all the vendors on board?"

37FEsoaevolve_in2.gif
Click for larger view.

It’s tempting to interpret the tension between these two perspectives as a replay of the cathedral and the bazaar — or perhaps instead, WS-Heavy and WS-Lite. In that dichotomy, WS-Heavy would refer to the security, reliability, and scalability that WS-* claims to deliver, whereas WS-Lite would mean the speed, simplicity, and agility that attract labels such as REST, AJAX, and RSS. None of the enterprise architects we interviewed for this story has pledged allegiance to either of these camps, though. They’re intensely pragmatic people who will do whatever it takes to get the job done, and it’s instructive to learn how they are -- and are not -- making use of Web services standards.

RouteOne: Securing Credit Checks

Although end-to-end SSL is often sufficient, RouteOne’s Subramaniam has two reasons to prefer the more granular approach enabled by WS-Security. First, it’s necessary to digitally sign the credit applications his application transmits, and to do so according to rules understood by service partners. WS-Security defines such rules, although admittedly, and unfortunately, too many of them. One method is to put the signed application into the body of the SOAP message; another is to use SOAP with attachments. In the end, there was no agreement among the service partners, so RouteOne uses both. That’s frustrating, but Subramamian would rather have two rules than none.

The second reason touches on one of the deep principles that motivates the design of the WS-* stack: pervasive intermediation. RouteOne is required to maintain meticulous audit logs and would prefer not to have to encrypt all of them. So it’s using DataPower’s XML router/accelerator to selectively encrypt only sensitive items such as gross pay and Social Security number. Because it’s a standards-based intermediary, the DataPower box can straightforwardly modify RouteOne’s XML message traffic in this way, and it could be swapped out for another appliance that did the same thing.

When services communicate directly, as many if not most still do, there’s no need to define the rules of engagement that enable service intermediation. Today’s most visible exemplars of WS-Lite -- Amazon and eBay -- use Web services in a point-to-point way. In that mode there’s not much difference between SOAP/WSDL APIs and REST APIs, so it’s not surprising that developers who work with these platforms overwhelmingly prefer the REST flavor. But when you do need to flow your XML traffic through intermediaries, SOAP and WSDL suddenly make a lot more sense.

Subramaniam is a pragmatist, however. Plain XML over HTTP, sans WSDL, also plays a role in RouteOne’s internal and external affairs. Because it’s a no-brainer to put a servlet interface onto an internal legacy system and pull XML data through it, that strategy is used where appropriate. Some of RouteOne’s external partners use the same approach, and because “they’re making money hand over fist” doing so, Subramaniam can’t mandate otherwise. Instead, RouteOne normalizes inbound traffic to SOAP and WSDL in order to enable its expected future use of BPEL (Business Process Execution Language) for service orchestration. Today, partners who don’t present SOAP and WSDL interfaces are not competitively disadvantaged. But the tipping point may not be far off.

RouteOne depends on both SAML and WS-Security, and Subramaniam wishes he could use a standard form of reliable messaging, too. “If I don’t send a message, we are losing money,” he says. Drawing inspiration from ebXML (e-business XML) and JMS (Java Message Service), he specified -- and is now using with partners -- a scheme that guarantees orderly and reliable delivery of messages. But he’d rather it were otherwise and hopes that OASIS will succeed in merging the two proposals it is now hosting: WS-Reliability and WS-ReliableMessaging. This duplication is “really, really bad,” Subramaniam says. “I wish we had a common spec so I could dump my stuff and just use it."

Corillian: Point-to-Point Simplicity

Many service-oriented systems don’t require reliable messaging and, according to Scott Hanselman, chief architect at Corillian, his company’s banking middleware falls into that category. Corillian’s product, called Voyager, handles services touched indirectly by 25 percent of all users of online banking, Hanselman says. “But the only transaction they care about is the one at the host.” So he’s not worried about the merger of WS-Reliability and WS-ReliableMessaging. Although he does make use of WS-Security, he regards SSL as equally effective in most cases. That approach precludes routers and intermediaries, he admits, “but rarely do I use them, because nine times out of 10 we’re doing point-to-point messaging.”

He’s also dismissive of UDDI, the much-maligned standard for publishing directories of Web services. What about the argument that services not  found in the yellow pages won’t be reused? Hanselman doesn’t buy it. Finding services isn’t really a problem for developers, he says. Using them easily and effectively is. Imagining a fictional average developer named Mort, Hanselman opines that SOA will be a nonstarter until we can shield Mort from XML angle brackets and X.509 certificates. To that end, he thinks the most important standard is WSDL because it’s a tool-enabler.

Of course WSDL has earned its fair share of criticism, too. RouteOne’s Subramaniam thinks that the “goofy” complexity of WSDL 1.1 made it a ball and chain that SOA has had to drag around, and he hopes that the “much cleaner” WSDL 2.0 will lighten the load. Perhaps, Hanselman says, but “you can’t unring the bell.” Millions of Web services transactions ride on WSDL 1.1 and will for a long time to come. Using WSDL 1.1, Corillian was able to describe the objects, messages, and services at the core of Voyager and to bind those descriptions to internal machinery that doesn’t speak XML. As the need arose, the company created alternate bindings that enable customers to see the engine through a Web services lens. If WSDL 1.1 was an 80 percent solution, Hanselman thinks, then WSDL 2.0 might be a 90 percent solution, but either can deliver crucial leverage.

37FEsoaevolve_in3.gif
Click for larger view.

Ohio State: Securing Vital Signs

The most widely adopted of the advanced Web services standards is clearly WS-Security. Beyond that it’s hard to find practitioners who have worked with the more exotic beasts in the WS menagerie, but Furrukh Khan -- who holds joint appointments in the colleges of engineering and medicine at the Ohio State University Medical Center and is broadly responsible for its medical IT -- tells a fascinating story about his transition from basic to advanced Web services.

In this scenario, vital signs flowing from monitors are recorded in databases and are simultaneously delivered to smart clients that observe, replay, analyze, and annotate the streams of data. The streams must be delivered to a lot of clients reliably, securely, and in near real time.

A first implementation, based on Microsoft’s WSE (Web Services Extensions), made use of WS-Policy, which hasn’t yet found a home in a standards body but likely will soon. WS-Policy was used to declare the means of authentication to back-end databases — for example, to require X.509 certificates signed by a specified key — as well as the required payload signature and encryption.

The current implementation -- based on the beta version of Microsoft’s Indigo, a Windows implementation of a stack of advanced Web service protocols -- uses WS-ReliableMessaging to ensure orderly and reliable delivery of messages. And it uses WS-SecureConversation to optimize that secure, reliable channel for high-volume traffic.

Khan explains that WS-Security alone, in concert with WS-Policy, could not sustain near-real-time traffic. The protocol, which required frequent exchanges of credentials with the identity management system, was too chatty. WS-SecureConversation, which enables caching of credentials, streamlines the protocol. That, coupled with a feature of Indigo’s implementation of WS-ReliableMessaging that enables a router to broker a connection between two end points and then get out of the way, resulted in a massive scale-up.

“Before, with WSE, each router limited us to 300 clients,” Khan says. Indigo can support 638 clients per router, he adds, and with optimization, that many clients for each service running behind the router. “So if you keep on adding services, it scales linearly,” he says. The system currently supports more than 1,000 clients, all observing vital signs simultaneously every 30 seconds.

Reflecting on the transition from WSE to Indigo, Khan echoes Scott Hanselman’s point about shielding developers from XML. WSE handled the basic scenarios, he says, but beyond those, “we had to go into the schema and do all the angle brackets.” Thanks to Indigo’s higher level of abstraction, that problem vanished.

More broadly, Indigo made a harder problem -- the appropriate use of Web services in concert with platform-native services and transports -- tractable. “Behind each Web service there’s an MSMQ [Microsoft Message Queue] and an enterprise service,” Khan says. “In the Microsoft domain, enterprise services are completely different from Web services, MSMQ lives in its own world, and XML has its own toolset.” Different team members had to be experts in different disciplines; no one person could master them all. From Khan’s perspective, Indigo gives “Mort” the leverage he needs.

Providence: Enforcing Contracts

1 2 Page 1
Page 1 of 2