Let me start with a disclaimer: This column is not intended to bash Web Services. Far from it. In fact, as I did my research, I became more and more impressed with how far Web Services have progressed, and the article sort of evolved into a list of things that, in my opinion, need to be fixed before Web Services will reach their potential.
Note that this is quite different from my stance on Enterprise JavaBeans (EJBs). Almost from the start, it's been my position that this is both a bad architecture and a solution without a problem. And yes, some of the issues that caused me to come to that conclusion about EJBs are present in today's Web Services architecture. But the difference is that Web Services for the most part only try to define a technology, whereas EJB tries to define your application. I'll talk about that a little more later.
This article addresses the following issues:
- What are Web Services?
- What are the alternatives?
- Benefits of Web Services
- Web Services architectural issues
- Web Services standards
- Web Services networking issues
First Things First--What are Web Services?
This question is actually larger than it looks. Understanding Web Services requires not only a technical viewpoint but also a business viewpoint, and in this case, the business angle is potentially much more revolutionary.
The technical side is definitely evolutionary in nature. All that a Web Service does is provide a standardized way for one program to call another. Despite all the bells and whistles and 400-pound specification documents, when you boil down the actual workings, what you get is a way for a program to send a request to another program and get a response back. All of the complexity centers on data formats and the like. How do I specify a complex structure? How do I send binary data? All of that is covered in the specification. In a nutshell, Web Services use HTTP as the transport mechanism to transfer Simple Object Access Protocol (SOAP) messages between two programs. I'll talk about the pros and cons of the issue later. Just remember that the basic idea is that if you can use a browser to talk to a machine, you ought to be able to access Web Services on that same machine (this is not always the case, but that's the idea).
The revolutionary part of this whole business is that two enterprises can actually find one another in cyberspace and join forces. When used internally on a LAN, Web Services are sort of like suspenders and a belt: redundant. A large part of the overhead of a SOAP message is due to the fact that an external source helps define the message. In an intranet application, you had better already know what the message contents are, and since you do, you can use a much leaner alternative, such as XML-RPC or RMI (I'll touch on those momentarily). But the promise of Web Services is that these services can be provided to anyone who wants to use them. By publishing the definition of my Web Service in a global repository, theoretically anybody who wants to can then go to that repository, download the definition, and configure a program to use my Web Service. Already, there are registries out there that provide literally hundreds of different services, from stock quotes to airline reservations to credit card validations. As an application integrator, you would use a search capability at one of these repositories, find such a service, and then, rather than actually write software yourself or integrate a third-party package, simply make a call to this Web Service.
This functionality does not come without a price, of course. You need to add two new standards to the mix: Web Services Description Language (WSDL) is the language used to define a Web Service, while Universal Description, Discovery and Integration (UDDI) is the standard used by all directories of such WSDL information. The idea is that you access a UDDI directory and retrieve a WSDL definition of a Web Service. You then incorporate that (typically using a simple wizard in your IDE) into your application environment, and you're ready to go.
Unfortunately, the revolution is taking longer than the evolution. There are a number of business concerns, including costs and security, but there are also a few technical glitches, and that's what I'd like to address.
What Are the Alternatives?
I already touched on a couple of alternatives. Many exist, especially for intranet solutions, where you have more control over both sides of the conversation. Remote Method Invocation (RMI) is one of the most straightforward solutions for Java applications. Essentially, you design your object in such a way that an application running on one machine has a "stub" version of the object that passes requests through to a real object on another machine. The primary benefit of this approach is that it's nearly transparent to the programmer. With a little setup, the application is pretty much oblivious to the communication.
Other options include multi-lingual distributed object techniques, such as Common Object Request Broker Architecture (CORBA) and Distributed Component Object Model (DCOM). For lack of a better distinction, DCOM is the Microsoft standard, and CORBA is everybody else's. You may have heard of CORBA under its Internet-enabled name, Internet Inter-ORB Protocol (IIOP). Whereas CORBA was originally more LAN-oriented, IIOP is CORBA over the Internet. In any case, these architectures allow compiled objects to communicate with one another. They arose primarily in response to the needs of embedded documents--a word processor needing to display a spreadsheet, for example. They worked best in homogenous environments where the machines all spoke the same language, at least at the operating system level. The problems arose when attempting to use these same technologies to communicate between disparate hardware and software platforms. CORBA, for example, has a nasty tendency to have vendor-specific proprietary extensions, making one CORBA incompatible with another, which defeats the purpose of inter-machine communication.
Change the Message--Using XML
Each specification has its own communications definition language, with no clean way to get them to talk with one another. What was needed was a common way to represent any kind of data. That's where XML comes in. Once XML was standardized, it became sort of the Esperanto of inter-machine communications. Finally, there was a standard that could be used to wrapper information inside of tags. A couple of communications protocols quickly came to being, the first of which was XML-RPC. XML-RPC is still around; basically, it consists of sending an XML document to a server and getting an XML document back. There are a few formatting rules, and it's not sufficient for complex messages, but it's a very straightforward protocol. It was designed to allow one program to call another over the Internet--nothing more or less.
The other major XML-based protocol is SOAP (in fact, you can think of XML-RPC as a sort of "SOAP lite"), which is the basis for Web Services and in my opinion one of its major problems. SOAP is the 800-pound gorilla of communications specifications. The SOAP specification is dozens and dozens of pages long, as compared to the XML-RPC specification, which is considerably shorter. By now, SOAP messages can include just about any kind of data in numerous formats. The SOAP standard went from really large in Version 1.1 to gargantuan in Version 1.2. And the SOAP committee isn't particularly hasty, either; it took three years to get 1.2 to the recommendation stage.
Change the Messenger--Other Transport Mechanisms
And if that isn't confusing enough, there are some emerging protocols that could vie with HTTP. The one I've heard the most about is Blocks Extensible Exchange Protocol (BEEP). BEEP allows you to establish a connection between two machines using XML and then transfer whatever data you want. (This could be a practical answer to the overhead problem I will be complaining about shortly.)
Benefits of Web Services
I promised I wasn't going to just bash Web Services, so let's look at the benefits. The primary benefits are that you can send just about any type of data and that it uses standard HTTP protocol. The communication is done via port 80, so, theoretically, you should be able to use it even through an HTTP server proxy, although as I'll explain later, that's not necessarily the case.
The SOAP specification is now transport-neutral, so SOAP messages should be able to pass through just about any messaging interface. I'm still a little unclear about why this is a good idea, but it's a huge issue for the designers. In theory, I can send a SOAP request to someone via HTTP, but it might pass through an MQ/Series tunnel to get there, and the MQ/Series tunnel might actually do part of the processing. I'm still waiting to see how that will all work together.
The biggest advantage, though, of Web Services is the incredible deluge of support from tool vendors. Every modern Web application design tool has some sort of Web Services tooling available. Our beloved WDSc is no exception; in fact, the Web Services support in WDSc is quite good. I managed to create a Web Service and get it running almost immediately.
Web Services Architectural Issues
The biggest issue is that SOAP requires tons of overhead to process short messages--for example, some 700 bytes for a typical Web Service SOAP request! Thus, it limits the applicability of the protocol. It's one thing if you are using a Web Service strictly to handle transaction-level processing; the overhead is probably unlikely to be significant (although a few milliseconds here and there tends to add up). However, by their nature, Web Service requests aren't the kind of thing you want to invoke in response to a user keystroke. Also, unless you're on a high-speed connection, response time will become an issue.
Next is readability. The entire Web Services specification definitely looks like it was written by committee. The WSDL is verbose and inscrutable, while at the same time providing odd defaults to attributes that should not be defaulted. I tried to debug a Web Service but found it very difficult to dig through the complexities of the interface. The generated code that comes from WDSc is so dense that I was able to modify only one part of the application.
Thus, the complexity of the interface has gotten so bad that human beings can't readily write the necessary code. I've heard it said that the WSDL files used to define a Web Service are "not meant to be read by humans." I'm uncomfortable with that state of affairs.
Web Services Standards
They beauty of standards is that there are so many to choose from. When you create a Web Service, you need to choose one of four standard styles: document/encoded, document/literal, RPC/encoded, or RPC/literal. Each style has its own strengths and weaknesses--except for document/encoded, which seems to not be used.
The style used most often is Microsoft's undocumented "wrapped" style, which if I'm not mistaken is also an option on the WDSc Web Services wizard. So, with all of the lip service paid to standardization, it turns out that one of the more popular styles is actually non-standard. Evidently, there weren't enough standards to choose from.
Web Services Networking Issues
A Web Service is supposed to be a remote procedure call interface using XML (well, using SOAP) over HTTP. So, for the vast majority of messages, you're just sending plain ASCII text over port 80, which should be about as network-transparent as you can get.
Except for one thing: It seems that Web Services and HTTP proxies don't always play nicely together. This is because not all HTTP proxies support something called PROXY-AUTHENTICATE, and that feature is crucial to forwarding Web Service requests. IANANE (I Am Not A Network Engineer), so I can't give you much more detail than that, but I can attest to the fact that HTTP proxies do cause Web Services to fail in ways that the LAN guys are unfamiliar with.
The big problem is that you probably generated your client by running a wizard of some type over a WSDL file, and therefore you have no real way of telling what kinds of errors occurred. All you know is that it is not working, because the Web Service sent you some cryptic error message telling you so. The black-box nature of the code makes diagnosing problems really difficult.
What Does It All Mean?
So, does that make Web Services bad? No, not at all. As I said at the beginning, I'm impressed by the number of Web Service providers out there already. Obviously, despite the glitches, a lot of people are willing to wager that Web Services will take off. And unlike EJB, the Web Services technology isn't flawed from the get-go; it provides a real service. The question is whether the standards committees can shake the bugs out before the proprietary versions take hold and strangle the standard.
But maybe that's just what Microsoft wants to have happen. I saw an interesting comment on a mailing list: "Maybe Microsoft has changed their strategy from Embrace, Extend, Destroy to Embrace, Glut, Destroy." That's something to keep in mind: Since we're using wizards to do all the work, are the Web Services "standards" starting to creep out of control? Just a thought.