Tom Housel
Professor
Naval Postgraduate School
Monterey, California
Deja Vu All Over Again
Congratulations to American enterprise. According to your cover story ("The Meter System," Summer), they have rediscovered the remote service bureau and time-share computers. Think Infonet at Computer Sciences Corp. in the '60s.
Andrew H. Olson
Managing Director
TEAM International Group
Gainesville, Florida
Utility Computing Is a Numbers Game
Norm Alster's excellent cover story sets the right tone for a review of utility computing with examples from the venerable history of the IT hype machine. In these volatile economic times, converting IT costs from fixed to variable sounds like a winner, but the reality is more complex.
Decades ago, in the premicroprocessor mainframe days, utility computing (in the sense of time-sharing on a mainframe) was the only game available to many companies because of system costs and the limited number of people who could operate or program such systems. Today the motivation for utility computing (in the sense of running applications remotely on someone else's computer system) has more to do with excess computer capacities (especially higher-end servers) and the differential margins on servers: larger servers command higher margins than PC-based systems, but few companies need the computing capacities of the higher-end machines and are reluctant to pay for them. Hence, large service-oriented vendors (IBM, HP) are promoting the concept of pay-as-you-go computing to both prevent margin erosion and capture market share.
Grid and utility computing are not identical. Grid computing involves distributing the load for a computational problem to multiple independent computer systems on a network (internally, externally, or both); the effect is to pool the resources of many machines to solve a problem (strength in numbers and all that). Grid computing was inspired by the needs of various national laboratories, such as Argonne and others: they wanted (and want) to attack "grand challenge" problems (e.g., protein folding) without each lab having to buy its own mammoth systems. In that sense, grid computing is both a funding opportunity and an interesting technical challenge. In the business world, grid computing is a response to the realization that all those PCs idling on desktops (especially at night) are an underutilized asset that with decent system software can reduce IT upgrade costs.
One key problem for utility or grid computing entails the cost of moving lots of data over a wide-area network. If large volumes of data must be moved over a WAN with a fast response, then the price of even minimally adequate network capacity will be high, perhaps costing more than the computer system over a couple of years. Vendors of utility computing that rely on high-volume data traffic over a WAN must factor in these costs, including how variation in the load can affect pricing. Unfortunately, competition in WAN service is not as fierce as that among computer-system vendors.
Interest in utility computing is motivated by the costs involved in the seemingly constant need to adapt one's IT infrastructure and applications to changing business processes. Yet outsourcing one's applications to a vendor's remote data center usually means sharing standard applications on its servers. In most cases, the vendor cannot or will not customize the application to meet individual customer's business-process needs. This lack of adaptability can be an ironic limitation to utility computing since nearly every package needs to be customized for the sake of organizational efficiency and competitive advantage.
Before opting for utility computing, consider the trend in computer technology. Over the past two decades, the price/performance of computer systems and components has dramatically improved (especially PCs). These factors have led to a world awash in computing capacity for most business applications. The physical size of systems has reduced space and HVAC costs as well. Application software is increasingly powerful, systems are easier to manage, and there are many more knowledgeable IT people than two decades ago. CFOs need to factor in the price/performance curve of systems when evaluating the possible cost savings from utility computing and whether their computational needs are rising faster than the computational price/performance curve is improving. If so, then utility options should be carefully examined, especially the service-level agreement. Often outright ownership is the more cost-effective route.
Tom Shillock
President
M2 Consulting
Portland, Oregon
Reader Comments» Post a comment