Introduction to the HP Virtual Server Environment

Chapter 2: Why Should an IT Organization Care About
Virtualization?
Virtualization has become a hot topic in the last few years. In the late 1990s, many companies were
growing very fast and it was very common for IT organizations to deploy separate servers for each
application that was required by the business. The end result was a sea of servers in most
companies’ data centers – most of which were seriously underutilized. Let’s explore how and why
this overprovisioning occurs and how virtualization can help to significantly reduce computing costs.
Overprovisioning
Servers are overprovisioned for a number of reasons. Some of these include:
• Responsibility of application groups for their own IT resources
• Variations in application load
• Difficulties with capacity planning
• Severe penalty for underprovisioning
Responsibility of Application Groups for their Own IT Resources
This is probably the primary reason why servers are overprovisioned. It is common for each
major application or business unit to have a separate application group that is responsible for its
own IT resources. This situation leads to overprovisioned servers because each server must be
sized to handle peak loads without regard to how much time the servers are underutilized. In
addition, the resources are usually dedicated and cannot be shared.
Variations in Application Load
Most data centers have two common types of workloads: applications that users interact with
each day, and the batch processing that is used to extract and process information from large
amounts of data, which typically occurs during off hours.
Both of these workload types differ in their load profiles. Interactive workloads tend to vary in
short windows of time. A workload might consume 15% of a system’s resources at one time and
then 75% a minute later. Batch processes typically have longer peak periods, but they tend to go
from 0% to near 100% and then back to 0% as the job starts, runs to completion, and then
finishes.
Because servers normally are sized for the peak load (and sometimes for additional growth), the
average load is usually low in relation to the size of the server.
Difficulties with Capacity Planning
Another reason for overprovisioning is that capacity planning is an inexact science. Customers
are attempting to predict the future using, in many cases, no real information about the resource
requirements of the workload. Even when real data is available, customers speculate as to the
amount of resources required on a new server, as well as expected growth of the workload over
the life of the system. The result is that, just to be on the safe side, business units typically
overestimate their requirements.
5