I have had the opportunity to perform a few VMware Capacity Planner assessments over the years and I have been, more the most part, pretty happy with the process and the results of the reports. The assessment is really pretty straight forward. We had physical servers to the project, making sure we have proper permissions to perform all the tasks and then let the process run over an extended period of time. For the most part, this way of sampling over an extended time frame will give you a very good idea what can be virtualized and the number of hosts that will be needed.
However, performance based metrics over time from VMware Capacity Planner have drawbacks. I came across an assessment where the tried and true methods did not quite tell the real story about the physical server. In this specific case, the physical server had its peek mode of operation everyday from around 3pm to 6pm and was pretty much resting the rest of the day. It is this limited work time that really skewed the results of this host’s and at first glance, the assessment had me considering taking away resources for the physical host during the physical to virtual migration. Here is an example of the collection results (click the image to expand).
At first thought, if needed, I could add resources to the machine once converted but that idea quickly got squashed since the physical server was a Windows 2003 Standard 32 bit Edition with four processors and four gigabits of ram. That will leave these servers in an “ify” state and would be a part of the second wave of migrations with special consideration on the failback process, if it becomes necessary.
The one mistake that I had that is worth sharing is when you are doing some kind of capacity planning engagement, once you have a list of servers to assess, to ask questions about the servers.
- Does your client expect any of the machines being assessed to not be a migration to virtual candidate?
- Do any of the servers have special tasks they perform that would not line up with the way VMware Capacity Planner works?
- Wouldn’t application response time and transaction load be a better way to assess the environment?
Until something better comes out, we have to work with what we have and in the long run it is good to establish a baseline of questions to ask, based on your own and your peers experiences.