Category: Software

After all the things I’ve written about using wireless access in public places, I was really sad to see that Mobilestar is in danger of closing.

I was talking this morning to some classmates, working on a project that was trying to identify customers’ perceptions of Zipcar. One of the perceptions was essentially “It sounds like a great service, but I’m skeptical.” In trying to articulate what was behind this perception, I said, “The customer doesn’t want his heart broken by another dot-com.” I hadn’t thought of it this way before, but that’s a lot of what I’m seeing in business school now. Lots of cynicism, lots of verbal defensiveness. Wall Street is the same way–there’s no rational reason that Akamai, a great company with a great business plan and great prospects, is trading at $4.26 a share. We loved dot-coms and now so many of them are gone. It’s like a grieving process.

More thoughts shortly on the history of web services. I just need to find a better place to write.

---

History of Web Services, Part II

This is part two of my informal, inaccurate History of Web Services; part one was posted on Wednesday.

CORBA was probably the first credible enterprise scale mechanism to allow processes (applications) on different computers to talk to each other. Why hasn’t it taken over the world?

Characteristics of CORBA

CORBA, according to its caretakers the Object Management Group, was designed to provide “interoperability between applications on different machines in heterogeneous distributed environments and seamlessly [interconnect] multiple object systems.” The calling system did not have to be aware of the hosted service’s location, operating system, programming language, or anything else about the service except for the interface (in the sense of an API–the defined way to talk to the service).

Sounds great in theory. You could theoretically make just about anything a CORBA-compliant object–legacy databases, Perl scripts, Microsoft Visual Basic applications–and have it all work together.

So what’s wrong in practice? The CORBA spec requires an Object Request Broker (the ORB in CORBA) to reside at a known location that keeps track of where all the services are. And that’s potentially a huge performance bottleneck. CORBA isn’t dead yet, but if you were to write its eulogy performance would be the cause of death.

Microsoft’s first answer: DCOM

Microsoft in the mid-nineties announced that it had a better answer to the distributed computing problem than CORBA. It was based on its existing Component Object Model (COM) that allowed interapplication communication inside a Windows machine, but it now had features in the plumbing to allow COM calls to work across a network. DCOM (Distributed COM) was supposed to make calling remote application services as easy as making OLE calls in Windows.

No, seriously, that’s what they said. The sad reality, at least from the point of view of thisex-Windows programmer of enterprise applications, was that making OLE calls in Windows was never particularly easy–or reliable. Different computers might, or might not, have functional OLE subsystems. The application being called might respond so slowly that the call would time out–even if both were on the same machine. You can only imagine how well distributing that same architecture worked.

Any distributed code you want, as long as it’s Java

The Java community made their own play at a distributed computing paradigm, called Enterprise Java Beans. At least one critic has called EJB the only “detailed, practical” specification of CORBA services. In addition to the core CORBA capabilities, EJB offers security, persistence, concurrency, load sharing, and other “value added” features.

From a programmer’s perspective, EJB sounds like (and is) a great technology. The issues with EJB are more subtle. First, you have to use Java to use EJB. This isn’t necessarily a problem, as Java is a pretty universally available programming language, but it places more constraints on your application than CORBA per se. Second, your application has to be pretty tightly coupled to the particular beans that provide its services. This makes operation in intermittently connected environments dicey at best.

Web Services

So what does the Web Service paradigm offer that CORBA, DCOM, and EJB couldn’t? That, it would appear, is the $64,000,000 question. The new wrinkles are XML, HTTP, and independence from language, platform, and broker.

No Broker: Web services can (and in most cases, do) run without a central broker. This eliminates one perk of CORBA, not having to know where the service is, but I’d argue that’s a questionable benefit compared to the performance hit imposed by going through a broker for all calls.

XML: Web services generally presuppose that the fundamental language of interchange for data is XML. This assumption doesn’t extend across any of the other approaches, where XML support, if it exists at all, exists as a bolt-on.

HTTP: Web services generally use HTTP as the transport protocol. This is important because it enables making web services calls through firewalls. However, as HTTP isn’t a robust transaction protocol, this may impose performance hits.

Platform and language independence: Unlike EJB, you don’t have to write Java to get in the game for web services (at least, most implementations). Unlike DCOM, there’s no platform requirements.

I was talking to some folks this weekend about my work experience, and mentioned that my programming experience had been in client server systems. “Client server!” they said–“boy, I haven’t heard that expression in a long time.” Feeling instantly old.

So what’s the connection between client-server and this XML-RPC/SOAP/scripting/web services thing I keep writing about? Kind of an indirect parentage, actually. It’s useful to go back a few generations to get the background on where web services came from, technologically speaking.

History Lesson

A long time ago, computer programs were monolithic, in the technical sense of that term. They were one big chunk of code that could be accessed only one way and only ran on one machine. This was because the machines were usually so big and expensive that there weren’t many machines around, and thus there was no value in networking them.

Somewhere along the way, things changed. Pretty soon, people were interested in getting systems to talk to each other. And software had started getting more structured in the meantime. On the one hand, you had application programming interfaces, or APIs: formally defined ways to access the functionality offered by a program, whether an application or an operating system. On the other hand, you had object orientation–the concept that a given chunk of software should protect the data that it accessed and offer well defined methods to read and change that data.

Why was this important? APIs gave software developers a clearly defined path to add functionality to applications or to write applications for platforms. (By the way, the Mac was the first mainstream personal OS with a documented, rich API for writing applications.) And object oriented meant that you didn’t have to worry about some other chunk of the application randomly changing your data, making it easier for large numbers of people to work together on a software project. But we’re still talking about monolithic applications running on only one machine. The network, if it’s there, is slow and unreliable for most users, or else only connects largely heterogenous systems inside your own company.

Add Network, Stir Vigorously

Shift gears for a second. It’s the eighties. You have a big transaction clearing system for a bank. You want to set up a network of machines to allow people to withdraw cash from their accounts, at places and times that are convenient for you. But you’ve got a problem. Even if you wire up ATMs in all the states where you have branches, you still haven’t covered the people who travel out of the area where your bank exists. How do you get an information system to allow people to get their money regardless of whose machine they’re using?

What if the software objects could talk to each other over the network? What would need to happen to make that possible? Well, one program would need to know how to talk to another and what to do with what the other said. (Sounds like an API.) And you’d want to make sure that only your bank’s systems actually made changes to the data–other banks could make requests, but not actually directly change the numbers in your customer’s accounts–so you could assure your customers’ security and privacy. (Sounds like object oriented code.)

Expose the API to get to a software object and make it accessible over a network. That’s the story behind CORBA (Common Object Request Broker Architecture). Behind DCOM (Distributed Component Object Model). And, to a very simplified degree, behind web services.

So if we had CORBA fifteen years ago, why web services now? That’s, as they say, a whole ‘nother story–one I’ll try to write about tomorrow.