How does a CIO or IT director go about implementing an enterprise grid? Are the grid products showcased in television and magazine advertisements the answer?
With the promises of dramatic IT cost reductions, increased system performance and increased productivity, grid computing is an option that needs to be explored in order to answer these types of questions.
Smart Business asked Paul Singleton of Perpetual Technologies, Inc., about the benefits, challenges and misconceptions of grid computing.
What is grid computing?
The phrase ‘grid computing’ came about from discussions of how the accumulation and delivery of computing power to applications should function similarly to the delivery of electricity through an electrical grid.
Grid computing allows a company to share its data-processing workload among its servers, workstations and storage, as long as excess capacity exists on each of these resources. As workload increases on one server, any grid-related tasks on that server can be allocated to another, less busy server. A true grid can utilize resources regardless of hardware vendor or operating system. No proprietary tool is required for participating in or managing the grid.
How are grids helpful?
Grids allow companies to fully utilize their resources rather than having excess capacity. Grids take advantage of different peak-load times of systems, including differences in time zones. This resource sharing and increased utilization reduces the rate at which companies need to upgrade and replace hardware. Large, expensive servers can be replaced with commodity hardware. More work is done with less equipment, reducing administration and maintenance costs.
Improved resource utilization has another benefit. It allows companies to proceed with strategic initiatives, previously delayed due to the cost of acquiring new, dedicated hardware. In a grid environment, these new systems require less powerful hardware because they can use the excess capacity of existing systems. Likewise, the extra capacity on the new system can be used by existing applications.
What are some common misconceptions about grid computing?
A grid should not be confused with a cluster. A cluster is a set of computers that work together as if they were a single system, but each node belongs exclusively to that cluster. They are not added and removed dynamically. Clusters use vendor-specific cluster management software to coordinate their work, and the cluster administration is performed centrally with the vendor’s tools. A number of these clusters may exist within a grid, along with other stand-alone servers and storage devices, but the clusters themselves are not grids.
A product might make use of multiple servers, but if the product is vendo-specific, uses proprietary protocols to work, or is centrally managed by the vendor’s tools, then it is not a grid.
What are some of the challenges to widespread adoption of grid computing?
For commercial, off-the-shelf products, licensing is probably the biggest challenge. Traditional licensing models include licensing an application for every system on which it runs, or licensing based upon the number of CPUs in the customer’s server. These are not well suited for grid computing. Throughout the day, an application might run on hundreds or even thousands of processors spread across dozens of servers and hundreds of workstations.
Other mechanisms for measuring usage will need to be developed. Creating licensing models that are simple but still acceptable to customers and vendors could prove difficult.
Companies developing their own grid applications face several technical challenges. For example, in a traditional architecture, an application has fast access to the data it requires. Storage and data are shared in a grid environment; this poses a challenge in keeping data close enough to the application so that data access doesn’t become a bottleneck.
Grid security can be more complex than traditional computer security, especially in environments where workstations, as well as servers, provide computing resources for applications. Protocols such as ‘grid security infrastructure’ are being developed to address authentication and secure communication between grid components.
Grid deployment within a company might face internal political struggles, as department heads try to maintain control of their assets, fearing reductions in their head count or budget. These issues are similar to those faced by companies as they centralize data processing resources. As a company adopts grid computing, application users and their managers might also have concerns about sharing their server’s resources with the rest of the company, especially when it may impact application performance. In reality, a grid only uses the server when that server has excess capacity.
PAUL SINGLETON is a senior Oracle consultant for Perpetual Technologies. Reach him at firstname.lastname@example.org.