Grid not ready for prime time

Grid computing is no empty buzz phrase. In fact, it’s a very substantial approach for scaling and optimizing distributed hardware resources. Grids aggregate idle processor cycles, storage capacity and other resources throughout networks, thereby serving client applications with supercomputer-grade performance. Depending on how broadly they’re implemented, grids can extend dynamic resource brokering, parallel processing and load balancing to all computers on an intranet, extranet and even a portion of the Internet.

Over the past year, grid computing has become an increasingly prominent theme in the road maps of platform, tool and middleware vendors. In January, several vendors announced development of the Web Services Resource Framework (WSRF) specifications for grid interoperability. In March, the Organization for the Advancement of Structured Information Standards (OASIS) established a technical committee to develop WSRF into an open Web services standard. Then last month, close to 20 vendors announced the establishment of the Enterprise Grid Alliance (EGA) to create a grid interoperability reference model, and address security and other issues critical to grids in corporate server clusters.

So grid is starting to mature, as a market and an approach for distributed processing. But the road to maturity is long, and grid computing won’t be ready for enterprise prime time for at least another three to five years. Some significant milestones must be reached before corporations can take for granted the presence of a ubiquitous, platform-integrated, standards-based grid infrastructure.

First, OASIS’ WSRF technical committee must finish work on its specifications. Then these specifications must be ratified and adopted broadly by grid vendors, including the Globus Alliance, which provides the industry’s dominant open source grid tool kit. Other grid industry groups must coordinate their work to provide comprehensive grid reference models, reference implementations, best practices guidelines and interoperability events. All this could take two to three years, considering the complexity of WSRF’s diverse specifications and the need to square them with other emerging Web services standards in areas such as security and management.

Furthermore, platform vendors must integrate grid features and standards natively into their products. Today, many production grids are implemented with Globus’ open source tool kit or with point products from small vendors. Currently, some platform vendors – most notably, HP, IBM, Oracle and Sun – have strong grid solutions and directions. But several other important platform vendors – most notably, Microsoft, Novell and BEA Systems – lack grid functionality and have yet to announce commitments to grid-enable their products. For example, grid is mentioned nowhere in Microsoft’s Longhorn wave for 2006/2007; consequently, the vendor probably won’t address grid in its core platforms until 2009 or 2010 at the earliest.

Finally, grid won’t truly mature till it breaks out of its traditional niche: serving the massive parallel processing needs of supercomputing applications in scientific and engineering environments. Grid potentially could be used to scale and accelerate all manner of applications, including search engines and application, database and mail servers. Oracle and IBM are ahead of the other platform vendors in this regard, having grid-enabled their respective application servers and, in Oracle’s case, its database and portal servers.

For the rest of this decade, grid computing will deepen its presence in its traditional scientific and engineering niche. However, grids increasingly also will penetrate a broader range of commercial environments, thanks to new standards and vendors’ growing commitment to this powerful paradigm.