Is grid computing the future? Currently, Microsoft doesn't have a grid-computing story to tell in the SQL Server space. But Oracle has been talking nonstop about the grid-computing capabilities of its new flagship database, Oracle 10g, scheduled for release next year. The potential performance and usability benefits of grid computing are hard to ignore.
What is grid computing? According to www.gridcomputing.com "computational grids enable the sharing, selection, and aggregation of a wide variety of geographically distributed computational resources (such as supercomputers, computer clusters, storage systems, data sources) and presents them as a single, unified resource for solving large-scale computer and data intensive computing applications (e.g., molecular modeling for drug design, brain activity analysis, and high-energy physics)." Picture an electric power network: multiple generators produce the power, but users access electric power without worrying where it comes from.
Theoretically, grid computing can bundle a series of inexpensive dual-processor servers into a powerful 8- or 16-CPU grid. Today, buying eight dual-processor servers is less expensive than buying one 16-CPU server. Is your 4-CPU database server running out of power? No problem--just add a new box. The grid will distribute the power and manage resources for you. This method of scaling out is in contrast to the scale-up idea (i.e., single-server SMP scalability), which entails adding processors to the existing box.
According to Oracle's Web site, "grid computing reduces the cost of IT by clustering servers together to act as a single large computer, dynamically shifting server resources between applications on demand." Oracle asserts that grid computing gives you high performance and reliability at a low cost. However, hardware and software limitations give grid computing more sizzle than meat for now.
When SQL Server 2000 first shipped, Microsoft hyped scale-out solutions as the answer to all high-end scalability questions. They backed away from scale-out solutions (in favor of single-node scale up) when they realized that reliable scale-out solutions might be more difficult to achieve than they'd originally anticipated. Database grids are more complex to create than grids for other types of applications because locking, transaction control, and state management are hard to manage efficiently across a grid of database nodes. Even so, I still believe that scale out will become the norm in computing and database processing.
In the meantime, I've seen a number of analyst quotes suggest that Oracle 10g's grid capabilities will be more "marketure" than architecture in the initial release. True grid computing for database users is still some time off. However, perception is often more important than reality when it comes to selling technology solutions, and Oracle will be aiming for the market-leader position--whether grid computing technology is ready for the real world or not. Because the problems that remain aren't trivial, I suspect that Oracle 10g's release won't set the world on fire. In any case, SQL Server users will be faced with hard choices if Oracle delivers a commodity grid-computing solution years ahead of Microsoft.
For more information about grid computing, visit the Global Grid Forum (GGF) at www.gridforum.org . The GGF is a community-initiated forum of individual researchers and practitioners working on promoting and supporting the development, deployment, and implementation of grid technologies. This site includes several interesting links and articles.