Grid Computing: Exploring the 'Tight-Coupling' Approach

By Stephan Schambach

Grid computing, long associated with high-end applications--including university and pharmaceutical research, engineering design and the like--has been gaining acceptance in some mainstream commercial businesses. But until very recently, it was not a factor in the e-commerce world. Now, a type of grid computing using tightly coupled resources is emerging as the basis for providing online shopping sites with critically needed response and capacity improvements. Such improvements will be particularly helpful in eliminating the problem of downtime during seasonal and other peak times.

The new approach to grid computing represents an important advance for online merchants who have been grappling with the skyrocketing numbers of people now making purchases on the web. All too often, because their online shopping sites simply cannot provide sufficient response time, the user experience suffers. And, unfortunately, if sites are not responsive to customers, e-tailers will not get a second chance to make a good first impression. Now, solutions based on tightly coupled grid computing are successfully addressing these issues by ensuring that retailers have the capacity they need when they need it without having to pay for peak capacity during non-peak periods.

WHAT IS GRID COMPUTING?
In general, grid computing can be defined as the linking of small computers to create a single, more powerful computer. But within that realm, there are different approaches. The best known of these are loosely coupled grids comprised of coordinated independent computing resources located around the globe. This approach is suited for programs that run in batch mode rather than real-time and works well with the applications used in research, mathematically intensive calculations, specialized engineering programs and so forth. It also is proving effective for the needs of some traditional businesses.

Loosely coupled grids operating in batch mode are, however, insufficient to meet the real-time needs of transactional applications. But a new approach is making the benefits of the grid concept relevant for e-commerce. This approach employs tightly coupled computer grids for small sets of highly orchestrated computers. These grids consist of worldwide points of delivery (PODs) comprised of high-capacity computing resources that are tightly architected to work well together and are managed by the service provider. Each grid is specialized for the delivery of specific applications and the platforms they run on. And, ideally, each is located in data center facilities that offer advanced physical security, power systems, environmental controls and Internet connectivity.

Unlike these new grid computing approaches, the traditional e-commerce solutions used by many online merchants today employ dedicated equipment for each customer. Dedicated equipment solutions are built to a specific capacity and trade off the cost of building too large versus the risk of not building enough should it be needed. Customers pay for peak capacity--even if it is needed only rarely and goes unused much of the time. For example, a customer who generally uses one unit of capacity may require 10 units at peak usage (e.g., during an e-mail campaign or unpredicted press coverage). With dedicated equipment solutions, that customer would have to pay for the full 10 units year round.

In addition, scaling to meet changing demands is difficult in computing environments with dedicated hardware and software architectures (since the components often are "hard-coded" to each other). For example, if the customer above had an additional spike in business and needed to upgrade beyond 10 units of capacity, his or her system might need to be re-architected, additional equipment purchased or contracts revised--leaving that customer unable to add capacity on-the-fly to meet changing demands.

And the capacity and response issues inherent with dedicated equipment do not end there. For instance, before instituting new campaigns, marketing departments always must be sure to inform the operations staff. This is necessary to ensure sufficient capacity, since successful campaigns have the potential to overload traditional e-commerce environments. These risks not only constrain marketing efforts, but also impact customer satisfaction in ways that are crippling to online business.

Tightly coupled grid computing solutions eliminate these problems by building scalability and flexibility into delivery platforms from the outset--thus enabling quick response to variations among customers and quick provisioning to accommodate those changes. With the ability to draw on the many pooled resources in the grid, these solutions offer sufficient capacity that can be allocated however, whenever, and to whomever it is needed.

MORE FLEXIBILITY
Optimized for e-tailing, these grids deliver on-demand applications as software and as a service in a whole new way. With the flexibility to respond to changes on-the-fly and pay-as-you-grow pricing, they can greatly improve response time and provide better user experiences. In a typical scenario, a customer who generally uses one unit of capacity may have an unexpected spike in activity on a particular day that requires 100 units. With grid solutions, customers can simply do whatever is needed to meet the demand (with no need to contact the service provider), and they will be billed for whatever extra capacity they use. Customers pay only for what they actually use (much as they do with utilities like telephone and electricity), while maintaining control over their sites.

Tightly coupled grid computing architectures also allow online merchants to run multiple, totally different look and feel websites in basically the same instance environment. From a system's perspective, all of the layers have separate permissions with no shared application handling multiple customers. Instead, the grid is logically partitioned at the user and operating system levels, so there is no application overlap between customers in the environment. Users get the isolation, but at the same time have full control over, and visibility into, the environment with the ability to leverage available resources to meet fluctuating demands.

These architectures offer greater redundancy and resiliency as well, in order to minimize or eliminate the impact of a component failure. For instance, if application or database servers, network switches or controllers in the storage subsystem failed, the POD would be resilient enough to quickly reallocate the resources needed for customer environments so that there is no disruption to commerce.

There is no doubt that today's online shoppers want what they want, when they want it--and merchants who fail to respond to those demands run the risk of losing out to competitors. Dedicated equipment solutions simply do not have the flexibility to respond quickly to such changing demands--and merchants can end up with downtime during critical peak periods. But with tightly coupled solutions that tap the tremendous potential of grid computing, online merchants will increase their chances of making good first impressions and keeping customers happy in the long run.

Stephan Schambach is the chairman and founder of Demandware, a Woburn, Mass.-based provider of e-commerce solutions that provide complete merchandising control. He can be reached at (781) 756 3700.