Introduction to the HP Virtual Server Environment

Batch Workloads
Batch-processing workloads have a very different resource profile. They tend to consume as much
resource as is available for the period of time they are running, and then drop back to zero.
These workloads also tend to result in a tremendous amount of unused resources because they
require resources to be allocated for the peak, even if the resources are not used while the batch
job is not running. The peaks are longer in duration, but the disparity between peak load and
minimum load is enormous. This situation is made worse if the server is dedicated to the batch
application and sits idle much of the time.
Other Costs Associated with Overprovisioning
Clearly, overprovisioning causes more money to be spent on server hardware, processors, or
memory. However, overprovisioning also results in other costs that are not obvious at first glance.
Direct Costs
Customers might be spending a significantly greater amount than is necessary for software
licenses. Most software licenses are based on the number of processors, so it stands to reason
that if you can reduce the number of processors, you can reduce your software license costs.
Having more servers or processors also results in higher maintenance and support costs, as well
as additional floor space, power, and cooling requirements.
Indirect Costs
Having more servers means that you need additional resources to manage them. Having fewer
servers and operating systems to manage allows you to free up those resources for more
productive tasks.
Key Components of an Adaptive Infrastructure
The traditional approach for dealing with varying resource demands was to size the server for the
peak demand. Sizing could then be optimized by controlling when certain workloads were running
to limit the possibility of resource contention. For example, interactive workloads were commonly run
on a system during the day, and batch processing was run during evening and night-time hours,
when users were typically not using the system.
Virtualization allows you to look at this from a different perspective. Rather than manage the
workloads to fit the resources, virtualization lets you manage the resources to fit the workloads.
Two key features are required to provide this different approach. The first is flexibility. Workloads
must be put into compartments that are flexible so that utility resources can be applied to the
workload when they are needed, and then are sent back to a free pool (or reallocated to another
workload) when they are no longer needed. The second is automation. It is not realistic to expect
administrators to recognize when resources are needed, identify where to take the available utility
resources from, and then reallocate those resources before performance problems impact workloads –
particularly with spiky interactive workloads.
Flexibility
Workloads must be placed into a compartment that can flex from the minimum fixed entitlement to
a maximum entitlement by acquiring additional utility resources. As discussed in later sections,
7