That being said, planning virtualization is a more complex process than most realize and too complex to fully detail here, but the key thing to understand is that planning for virtualization starts with a thorough understanding of the application portfolio:
- What is the criticality of the application? How will the risk profile change when consolidation is achieved (possibly requiring increased investment in fault tolerance)?
- What is the utilization profile? Is it enough to look at cpu cycles or will disk or network I/O be a factor as well? Can periods of high utilization be easily lined up with lower utilization periods for other applications?
- What happens when the application runs out of capacity? What kind of capacity planning function will you need to have in place to avoid resource issues?
- Does the current infrastructure present barriers, such as server-attached storage that will need be migrated to a SAN before the server virtualization can take place?
- Who does the application belong to? Depending on the organization (and the business unit), barriers to virtualization may or may not be raised by the users. Other possible conflicts related to ownership include location and regulatory compliance requirements.
- Is the application new, and does it need to be up and running in a hurry? Fast provisioning of resources is a clear strength of virtualization.
Typically an organization will start with development and test servers that have the advantage of random peaks and lower reliability requirements than the production boxes. As expertise grows they will start with production applications that have low overall and peak utilization numbers and that have low redundancy requirements. As savings materialize, and as new machines come in with the standard refresh cycle, some of the larger, more critical applications will be consolidated. Depending on what they are, the oldest, largest, most critical apps may never be virtualized, as the business may not want to risk adding additional complexity to a critical system that already works well.
This was first published in August 2007