The compute Grid is well understood today — but much less time has been devoted to getting data where it needs to be, when it needs to be there, as well as to how this effort is managed.
Enterprise users say the ability to manage data on Grids is a key requirement for accelerating Grid deployments within their IT organizations. Some large enterprises are finding that limitations in data management capability mean they must hold off on evolving their Grid deployments. Those that have moved forward have usually done it either through internal work, or through customized or cutting-edge offerings from vendors. Most early adopters have long-term plans to extend their activities from initial beachheads to multi-application and cross-organizational Grids. But without proper data management tools in place, applications will not perform well on top of a Grid architecture, and the cost and performance advantages of implementing a Grid will not be realized.
Commercial enterprise Grids require a data management infrastructure that allows end users and applications to share information, regardless of where it resides, and provides secure access to heterogeneous databases, middleware, file systems and storage systems. Traditional data management techniques are well established, but they were designed to run on centralized mainframe or client/server architectures and need to be adapted and extended for Grid architectures. If Grids are to progress into mainstream commercial use, a model for transactional Grids is needed that can support the kind of transactions that underpin commercial organizations. Increasing the availability of commercial applications for use on Grids is seen as key to driving accelerated adoption. Some combination of caching, data streaming, replication, global resource namespaces, data movement, data transformation, data quality and storage volume virtualization may be required, depending on the application and system architecture. As it stands now, no single approach — with the exception that a virtualized environment is necessary — or single vendor or group has a leadership position, and no one can address data management on every part of the stack. The challenge has been characterized variously as creating the data Grid, storage Grid, information Grid or integration Grid. The 451 Group believes the ability to manage data on Grids is the key to all of these.
Many enterprises, vendors and users have identified the transformation to a service-oriented architecture (SOA) as a strategic, long-term goal that can better align business with IT and improve responsiveness to changing conditions. Financial services companies, for example, see Grids as the underpinning for SOAs, which cannot be implemented without sophisticated data management techniques. A SOA uses short transactions and large volumes of associated data elements. For many organizations, SOA is the future for their enterprise IT environments. Grid computing is seen as the infrastructure model, and SOA as the application model. But SOA is not an exclusive role for Grid technology, which is also regarded as the underpinning for utility computing, a service delivery model. Equally important is how Grid technology relates to event-driven services, messaging, database systems, networking systems and legacy assets.
Grid middleware/scheduling vendors themselves have not ignored data management issues, especially as they seek to address a broader piece of the Grid “stack,” penetrate new markets and move beyond high-performance computing Grids. The problem is that data management is not part of their core skill set.
Platform Computing has some rudimentary caching capabilities in Symphony, but it typically partners for data management functions. DataSynapse has added data management virtualization functions to GridServer, which incorporates some of its distributed GridCache. The result is that more scalable, transactional applications that were once unsuitable for Grid deployment can now be run on GridServer. United Devices, perhaps the most unashamedly “compute-oriented” of this group, says it will partner for most data management functions and build some of its own, but it has yet to expand on this plan.
The major vendors — as always — can have a huge influence over technology directions; although they are not always working at the cutting edge of technology. Oracle's marketing of its database as 10g (the “g” is for “Grid”) has raised awareness of Grid technology to new levels, despite the fact that much of what Oracle is actually shipping is not viewed as a “real” Grid implementation by many. It is really database clustering. Oracle's view is that clustering is a good way to implement Grid capabilities — without heterogeneity. It sees customers moving from infrastructure consolidation projects and on to Grids. Typical offerings rely on federating access to other resources but not integration, although Oracle's answer is that customers should put it all in an Oracle database and then these concerns go away.
However, Oracle has done a good job of highlighting the lack of transactional support in Grids with its associated low-latency, high-volume data requirements, while other companies have yet to face the challenge — although startups such as CipherGrid are beginning to address this issue.
Microsoft has not talked much about its plans for Grid technology until recently, and it has confined the work it has done to the high-performance computing sector. But the signs are that Grid technology is about to enter the mainstream at the company. The launch of SQL Server 2005 later this year and Windows Server Compute Cluster Edition next year are the key events to look out for. In the longer term, Microsoft is aiming to provide what it calls the “unified Grid,” virtualizing the resources customers have in a heterogeneous environment.
The addition of Tony Hey, former UK e-Science initiative director, as a corporate vice president with Microsoft's technical committee, is another important indicator of Microsoft's future with Grid computing. With the Global Grid Forum having agreed on a vision for the Open Grid Services Architecture that does not mandate either IBM's WS-Resource Framework or Microsoft's WS framework for its implementation, Hey believes there is now a way for Microsoft to participate in the Grid industry. He expects Microsoft will play a role in open Grid standards and then will implement them into Windows.
The big server vendors all have plans for on-demand, utility or datacenter automation frameworks that will incorporate Grid technology as their underlying infrastructure. IBM views data integration as an important part of the Grid environment, relying heavily on information virtualization. IBM's vision is to virtualize all of the information in a distributed system — data sources from local or distributed file systems or databases/representations. To accomplish this, it would then wrap a global name around it and assign it a policy for how it is to appear. It is an aggressive technology direction, but IBM has invented or been responsible for many of the most advanced data management techniques.
Hewlett-Packard has plenty of Grid activities under way but hasn't made much noise about them, partly because it decided early on that Grid computing is going to be a horizontal, transformational technology that will need time to mature because of the profound organizational and cultural challenges it implies, and because of the time it will take to turn everything into a service.
Sun's Grid Engine distributed resource manager and workload scheduler remains the mainstay of its Grid middleware offering. More sophisticated data management capabilities, including additional caching and replication technologies, are scheduled to be included in phase three of its Sun Grid Storage Utility buildout, and Sun will either partner or acquire to obtain these.
A mixture of approaches — data movement, replication and data federation — will be necessary to handle the growing number of disparate data sources, including those outside of the database, as well as the growing number of devices that need to access them within enterprise IT environments. The 451 Group expects to see broader use of federated data access, distributed main memory and local disk caching techniques in future products. The 451 Group also expects to see increasing data management support for Grids embedded within application servers and databases. But because this implies a return to a more centralized application-server approach that does not fit easily into Grid architectures, an alternative development platform-oriented model supporting multiple programming language interfaces will also continue to be a requirement.
For more information about this topic, please visit www.the451group.com/intake/gridtoday-aug05.
About William Fellows
William Fellows is a principal analyst at New York-based The 451 Group — an independent technology industry analyst company focused on the business of enterprise IT innovation.