With the compute element reasonably well understood, the next challenges in Grid computing are essentially those that are “beyond the compute Grid.” The 451 Group has learned that many of the same key challenges and concerns are shared by early adopters from different vertical markets. Their concerns correspond to what have been identified by many early adopters that The 451 Group has spoken with as the main challenges — and barriers — to adoption.
Software licensing has become the key concern for an increasing number of early adopters. Put simply, for all the potential benefits of Grids, early adopters cannot afford to buy software licenses for every device in the Grid – – a necessity, as the Grid, by its nature, consumes resources dynamically. The 451 Group believes that as early adopters evolve into using Grids as a more mainstream technology, the restrictions of current software licensing will become an ever-greater obstacle.
Many of the Grid vendors talk about supporting the ability to proactively manage the use of software licenses based on business objectives. But early adopters with whom The 451 Group has spoken have their doubts; they want real and tangible changes. They see metered usage as a potential way to manage software licenses, but early adopters also say that even the offer of more flexibility in small things — like the rollover of unused license minutes and the separation of CPU/review and report time — would go a long way toward easing concerns.
The 451 Group believes that there may be an opportunity for more forward- thinking independent software vendors in each sector to gain advantage and steal market share by adopting more flexible models, but this hasn’t happened yet. Alternative purchase models (i.e., subscription, outsourcing, pay-as-you- go), as well as new technologies such as multicore and virtualization, suggest a change is under way that will have a cumulative and disruptive impact on vendor licensing policies and practices. Therefore, software licensing for Grids must be seen within the context of the changes taking place throughout the industry, as there is a lot of money at stake.
As the compute Grid is now well understood, much less time has been devoted to analyzing how data gets to where it needs to be, when it needs to be there, and to how this overall process is managed. As the momentum of enterprise Grid adoption builds, the ability to put data in the most suitable places, so that it can be shared with other applications when required, is increasingly coming into the spotlight. Industry leadership in this area is still up in the air.
In the pharmaceutical industry, for example, larger companies that are very geographically dispersed cannot take advantage of Grids over a wide area because of the problem of data management. They typically have large data sets rather than lots of smaller sets like other sectors. Although they can spray jobs across resources at different locations, some early adopters have said that the time it takes for the process to then locate the right database, extract the data, take it to where it needs to be and resynchronize — with the attendant latency and cost of bandwidth (building/leasing big pipes is expensive for commercial companies) — means this approach is simply not cost- effective.
The industry needs a common approach to getting data in and out of Grids, as well as standard I/O interfaces that application developers can work with. Only Oracle has done a decent job of highlighting the lack of transactional support on Grids and no one else has really picked up the challenge.
Cultural and Organizational
Another obstacle that early adopters face is the notion of cultural issues. Most early adopters have long-term and far-reaching plans to extend their Grid activities, but reaching beyond departmental deployments brings issues like trust, control, sharing and ownership into play. Specific issues range widely, from job fears to lack of awareness to more serious concerns, such as slightly different results from each job run on the Grid. Cultural resistance has also included interdepartmental issues and sharing of data.
Another issue is standards. When considering actual implementations, early adopters do not typically appear to care much about standards. But early adopters have also been vocal about wanting the standardization of the stack and APIs, like data input/output. They want only one set of standards and one stack, but the growing number of Grid industry early adopters suggests confusion and growing complexity in approach, not standardization.
There is a threat that the momentum Grid computing has gathered will be challenged by the inability of organizations to converge rather than collide. If Grids can find a place in one of the open source stacks — such as LAMP — it would certainly help develop or increase awareness.
Design Models, Development Approaches
At this point, only a few applications have been written specifically for Grid computing and only a small number of today’s applications have been deployed on Grids. The tools, models and infrastructure technology for developing Grid applications and services have been mostly focused on high-performance computing to date. But as enterprise momentum for Grid computing increases, new design points, development models and tools for creating and enabling commercial applications for Grids will be required.
The 451 Group believes that what happens in Grid computing market will depend a lot on what the major vendors do. A convergence between Grids and Web services will certainly help drive new approaches, especially if Grid computing continues to look like a good way to support service-oriented architecture (SOA), datacenter automation and utility models. But does this enhance the possibility of “Grids” becoming the next “solution”?
Enfranchising Current IT
The 451 Group’s research of Grid computing involves examining how Grids relate to SOA, event-driven services, messaging systems, database systems and networking systems, as well as how legacy assets and new technology interoperate. While it is a minor concern for early adopters, how and where Microsoft decides to participate in the Grid computing market will be important. Grid research at Microsoft has been a subset of its high- performance computing research efforts. But last year, Windows Server Compute Cluster Edition (formerly known as Windows Server 2003 HPC Edition) was moved over into the mainstream Windows Server development team, as Linux, distributed computing, clustering, Web services and 64-bit computing begin to drive commercial computing requirements in significant ways.
As a discrete packaged Windows release, CCE is expected to be the primary platform for future Grid introductions from Microsoft over the next few years. It will be focused on HPC clustering, MPI and jog scheduling. By version 2, which is expected in 2008, it will support resource scavenging on desktops and servers, with more data sharing functions for enterprises than the SETI model, according to Microsoft. Also, the Globus Consortium has pledged to support the creation of a Windows version of Globus Toolkit 4 (.NET and C#).
Coming Up Next
In the next Grid computing report (to be released in November), The 451 Group will look at utility computing. Grids are being used to underpin new generations of “utility computing” offerings, and there has been increasing interest in the development of outsourced or managed services using Grids. This report will examine the offerings and strategies of key vendors — including their pricing, metering and management mechanisms — and will assess from a user point of view what users’ needs are and whether they are indeed getting a “utility” service.
For more information, please visit www.the451group.com/intake/gridtoday-17oct05.
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.