Currently, most enterprise grid implementations reside in large enterprises like Hartford Life, Bank of America, and Royal Dutch Shell. The typical model is that of taking a complex computing problem, dicing it up into smaller problems, and distributing it to loosely stitched networks of grid-enabled computers.
Implementations of this type are what are termed “compute grids,” or linked servers and desktops that create what is essentially a virtual supercomputer.
Clearly, this model doesn’t lend itself well to broad adoption. The applications running on compute grids are often developed in house, and the grids themselves tend to be application specific and resource heavy.
According to Steve Tuecke, CEO of Univa, a grid software startup, organizations accomplish one thing when using compute grids: they take a single application and enable it to run faster by sharing computations across distributed resources.
“The problem is that you end up creating grid silos,” Tuecke said, “so you end up with an application that can run in parallel across different resources, but you still have a silo.”
Two other types of grids, data grids and resource grids, promise more flexibility.
Data grids distribute database information and storage, while resource grids enable the broad sharing of infrastructure, including servers, storage, and other data-center resources.
In order to stitch all of these disparate resources together, early adopters and grid startups are setting their sights on open standards.
The Red Hat Model
Univa, for instance, has adopted a business model that is, essentially, the same as the Linux vendor Red Hat. Whereas Red Hat distributes and services Linux, Univa focuses on Globus open-source middleware for creating grid networks.
The Univa grid suite includes software services and libraries for resource monitoring, discovery and management, as well as security and file management features. Univa’s business model makes sense, since its co-founders (Tuecke, Ian Foster, and Carl Kesselman) developed Globus in first place.
Univa got a big boost earlier this month when IBM licensed its software suite, using it both for eServer platforms and internal IBM grid projects.
“We will work closely with Univa on delivery of enterprise-ready implementations of Globus for IBM platforms in much the same way that IBM works with Red Hat and Novell to ensure Linux distributions on IBM platforms,” said Ken King, vice president of grid computing at IBM.
While Globus is currently the de facto middleware for grids — having the support of Cisco, IBM, Intel, Hewlett-Packard, Sun and others — not every major grid player has jumped on the Globus bandwagon.
Oracle, for instance, is pursuing its own, proprietary means of stitching data grids together, while Microsoft’s grid intentions remain fuzzy at best.
Another grid company, Data Synapse, takes a slightly different open-source approach. DataSynapse designs its products to conform to Globus and other standards, including SOAP and WSDL, but the company focuses on virtualizing business logic and data access.
“We virtualize application components,” said Jamie Bernardin, CTO and co-founder of Data Synapse. “Our approach is to grid-enable applications to perform as services. We shrink common resources down and make them available to other applications, with the result being that an organization’s applications are detached from specific hardware, connecting instead to an abstract layer.”
Is SOA the Answer?
As grid technology penetrates the enterprise, services are becoming the buzzword. Whether these are applications functioning as Web services or database resources acting as services under a utility model, the message is clear: IT executives want to utilize grids to make both their hardware resources and their applications more flexible and efficient.
Bernardin noted that while several standards are well suited to grids including Java, J2EE, WSDL, SOAP, and XML important pieces are still missing.
“You need standards for how applications communicate to each other,” Bernardin said. “Actually, we need standards for how to build grid-ready applications in the first place.”
Frank Gillett, principal analyst for Forrester Research, agrees that certain standards, most notably SOA (service-oriented architecture), will have an impact on how grid computing evolves since SOA defines how various computing resources interact with each other.
Under SOA, applications are built so that key components function as discrete services. Each service is decoupled from the larger application and functions on its own but, when taken together, this array of services adds up to a large application.
Obviously, this approach to application design is similar to the hardware design of grids. “Adding SOA-based Web Services to existing applications provides a translation layer,” Gillett said. “One application API is able to communicate with another application API.”
By utilizing SOA and Web services, data is freed up from one application and available to others. But that’s only one piece of the puzzle. When you start turning to standards as the keys for solving the grid-computing puzzle, which standards do you turn to?
“There is this giant alphabet soup of standards,” Gillett cautioned, “and to me, there’s no clarity as to which standards will matter.”
Highlighting his point, Gillett pointed to the many standards bodies relevant to grid computing: GGF (Global Grid Forum), OASIS (Organization for the Advancement of Structured Information Standards), SNIA (Storage Networking Industry Association), EGA (Enterprise Grid Alliance), and the IETF (Internet Engineering Task Force) — to name only a few.
“It’s too early to start betting on specific standards because there’s no agreement on which ones matter,” he said.
Even if grid standards are still a muddle, this much is clear: when early grid adopters look at grids, they eventually start thinking about Web services. How those two trends will converge, however, remains an open question.