Companies have long dreamed of assembling their enterprise systems from a collection of network building blocks. CORBA and DCOM, both early attempts at tackling this problem, never got very far in terms of adoption. But now that Web services have burst onto the scene, it looks like SOAP and WSDL will succeed in becoming the lingua franca for distributed computing, thereby providing the catalyst for a wholesale move toward service-oriented architectures (SOAs).

An SOA implements each part of a system as a Web service. Simple Web services provide low-level features such as access to a particular kind of data, and the ability to send out a piece of e-mail or perform a calculation. Higher-level Web services orchestrate lower-level services to provide more complex behaviors, resulting in a pyramid of increasingly specialized processing.

The big challenge that faces an SOA is how to organize these Web services into a coherent system, especially when they'll probably be implemented using a variety of platforms and languages. An SOA requires features like clustering, fault tolerance, load balancing, security, and management to work in a way that is independent of the services themselves, so application server technology isn't the answer. This is where grid computing, an up-and-coming concept, comes into play.

Grid-computing platforms operate in a manner similar to the national electricity grid, transparently connecting producers and consumers of services while shielding them from issues like failover, load balancing, and clustering. These platforms can link Web services from any vendor, and typically adopt a P2P architecture in order to operate on a large scale.

In the past, grid-computing platforms were positioned primarily as a way to make use of spare CPU cycles. But now that Web services have become popular, they're rapidly being viewed as a promising architecture for supporting large-scale SOAs. IBM's new CEO, Sam Palmisamo, recently said, "The grid is the ultimate method whereby you can establish this seamless, open-standard computing model." And Irving Wladawsky-Berger, one of IBM's top strategists, believes that "grid computing is really the natural evolution of the Internet," so there's obviously a lot of cool stuff about to happen in this area.

The fun part, of course, is figuring out good ways to implement a grid-computing platform.

One solution to the service lookup problem is to use a federation of UDDI servers as a kind of DNS for Web services. When a program requires a particular kind of service, it asks a nearby UDDI node for the location of a Web service that implements a specific WSDL interface. If that UDDI node doesn't know the answer, it delegates to a more knowledgeable UDDI node, propagating the request until an answer is found. Answers to Web service lookups can be cached at local nodes to accelerate subsequent lookups for the same kind of service. If a particular service fails, a client could automatically rebind to an equivalent service by repeating the lookup, assuming that the original service is stateless.

There is of course a lot more to a grid-computing platform than just locating and binding to services. For example, how are events reliably distributed to services throughout the grid? How are transactions conducted? And how do you manage an application whose services can be widely distributed among a large number of nodes?

Another deeper question is whether a grid-computing platform should connect producers and consumers of data as well as services. For example, a grid could allow a program to upload a document for use by other programs, automatically replicating the data to ensure performance and reliability, and allowing it to be located easily using a query language. The similarities between sharing of services and sharing of data suggest there might be a way to unify these two features and solve them with a single architecture.

The promise of grid computing is to provide simple, universal access to services and data with the quality of service associated with utilities such as electricity and telephones.

However daunting the technical issues might seem, I'm convinced that commercial service-oriented grid-computing platforms will be available soon. The metaphor is too attractive to ignore, and with the advent of Web services, the timing is perfect.