Web + Services

A lot has been written recently about Service-Orientation and Web services and REST and the massive amounts of confusion that seem to be surrounding the whole subject. After much navel-contemplation, I’m convinced that the root of the problem is that there’s two entirely orthogonal concepts that are being tangled up together, and that we need to tease them apart if we’re to make any sense whatsoever out of the whole mess. (And it’s necessary, I think, to make sense out of it, or else we’re going to find ourselves making a LOT of bad decisions that will come to haunt us over the next five to ten years.)

The gist of the idea is simple: that in the term "Web services", there are two basic concepts we keep mixing up and confusing. "Web", meaning interoperability across languages, tools and platforms, and "services", meaning a design philosophy seeking to correct for the flaws we’ve discovered with distributed objects and components. These two ideas, while definitely complementary, stand alone, and a quick examination of each reveals this.

Interoperability, as an idea, only requires that programs be written with an eye towards doing things that don’t exclude any one platform, tool or technology from playing on the playground with the other kids. For example, interoperability is easy if we use text-based protocols, since everybody knows how to read and write text; hence, HTTP and SMTP and POP3 are highly-interoperable protocols, but DCOM’s MEOW or Java’s JRMP protocols aren’t, since each relies on sending binary little-endian or big-endian-encoded data. Interoperability isn’t necessarily a hard thing to achieve, but it requires an attention to low-level detail that most developers want to avoid. (This desire to avoid low-level details isn’t a criticism–it’s our ability to avoid that kind of detail that allows us to write larger- and larger-scale systems in the first place.)

This "seeking to avoid exclusion" requirement for interoperability is why we like using XML so much. Not only is it rooted in plain-text encoding, which makes it relatively easy to pass around multiple platforms, but its ubiquity makes it something that we can reasonably expect to be easily consumed in any given language or platform. Coupled with recent additions to build higher-order constructs on top of XML, we have a pretty good way of representing data elements in a way that lots of platforms can consume. Does interoperability require XML to work? Of course not. We’ve managed for the better part of forty years to interoperate without XML, and we probably could have kept on doing quite well without it; XML makes things easier, nothing more.

Services, on the other hand, is a design philosophy that seeks to correct for the major failures in distributed object and distributed component design. It’s an attempt to create "things" that are more reliable to outages, more secure, and more easily versioned and evolvable, things that objects/components never really addressed or solved.

For example, building services to be autonomous (as per the "Second Tenet of Service-Orientation", as coined by Mr. Box) means that the service has to recognize that it stands alone, and minimize its dependencies on other "things" where possible. Too much dependency in distributed object systems meant that if any one cog in the machine were to go out for some reason, the entire thing came grinding to a halt, a particularly wasteful exercise when over three-quarters of the rest of the code really had nothing to do with the cog that failed. But, because everything was synchronous RPC client/server calls, one piece down somewhere on the back-end meant the whole e-commerce front-end system comes to a shuddering, screeching pause while we figure out why the logging system can’t write any more log messages to disk.

Or, as another example, the First Tenet states that "Boundaries are explicit"; this is a well-recognized flaw with any distributed system, as documented back in 1993 by Wolrath and Waldo in their paper "A Note on Distributed Computing". Thanks to the fact that traversing across the network is an expensive and potentially error-prone action, past attempts to abstract away the details of the network ("Just pretend it’s a local call") eventually result in nothing but abject failure. Performance failure, scalability failure, data failure, you name it, they’re all consequences of treating distributed communication as local. It’s enough to draw the conclusion "well-designed distributed objects are just a contradiction in terms".

There’s obviously more that can be said of both the "Web" angle as well as the "Services" angle, but hopefully enough is here to recognize the distinction between the two. We have a long ways to go with both ideas, by the way. Interoperability isn’t finished just because we have XML, and clearly questions still loom with respect to services, such as the appropriate granularity of a service, and so on. Work remains. Moreover, the larger question still looms: if there is distinction between them, why bring them together into the same space? And the short answer is, "Because individually, each are interesting; collectively, they represent a powerful means for designing future systems." By combining interoperability with services, we create "things" that can effectively stand alone for the forseeable future.

And in the end, isn’t that what we’re supposed to be doing?