Historically, IT organizations have relied on large, multi-CPU symmetric multiprocessing (SMP) servers for data
processing. The underlying assumption was that by adding capacity-more CPUs, memory, and disk-IT could
answer the need to process greater data volumes in ever-shrinking load windows.
That capacity, however, came at a high price. Acquisition, maintenance, and support of a single SMP server could
amount to millions of dollars. And SMP systems offered little flexibility to "scale down," meaning that costly
resources were often underutilized except for periodic peak loads. Faced with budget reductions in the early 21st
century, IT organizations began to explore alternatives for more cost-effective data processing platforms.
The grid computing architecture is rapidly emerging as a compelling alternative for data processing. A grid is
typically a collection of low-cost, commodity blade x86 processor-based server nodes connected over a high
speed Ethernet network in which resources are pooled and shared. Grid computing can offer several advantages
over monolithic SMP systems:
• Greater flexibility for incremental growth
• Cost-effective scalability and capacity on demand
• Optimized resource utilization
Despite its benefits, the grid computing paradigm presents a number of challenges to both IT organizations and
infrastructure vendors. Software applications running on the grid-databases, application servers, storage systems,
data integration platforms, and others-must be equipped with grid-specific functionality to take advantage of a
grid's capability to evenly disperse workloads across multiple servers.
