virtualization.info

As discussed in the first post, data is vital to being able to understand the opportunities in your environment and calculate important cost savings opportunities, without this data you cannot sell the solution and get buy in from upper management, which is key to your success. Therefore, data is the first category in this series.

Data and Assessments
There is no question that virtualizing servers saves money, putting several virtual servers on one physical server vs. several physical servers will show a save in many ways. Although, without data and metrics to prove the save to those sponsoring the efforts, the work is pointless.

Data is the key element to a successful virtualization program. Without data and metrics the successes of the program cannot be presented. Without data and metrics funding cannot be obtained to start or continue the program. What servers should be virtualized and why? Without the proper data and metrics that question cannot be answered.

In order to prepare a proposal to start a virtualization program the current physical environment must be assessed. An assessment can be done internally if the proper tools are in place, or there are many third party services today for virtualization assessments. Performance metrics of the physical servers are a must, and there are four primary utilization metrics that need to be captured:

• CPU Utilization
• RAM Utilization
• Disk I/O Utilization
• Network I/O Utilization

Make sure there is a standard and consistent model for how these metrics are obtained. For example, if metrics are captured 7 days a week and 24 hours a day, then the averages will tend be lower due to capturing both peak and non-peak metrics. Therefore a server running at 30% utilization for CPU on average maybe at 80% during peak hours and may not be a good candidate for virtualization, even though 30% seems like a low number. The key here is to be consistent in how data is obtained and assessed.

Beyond these utilization metrics, knowing the model of the physical server, the CPU speed and number of CPU’s, the age of the server and where it is in the depreciation cycle (if a depreciation cycle is used) are also key data elements of the physical server. For example, if a server is running at 70% CPU utilization it may not seem like a good candidate for virtualization. Although, if the server is four years old and the CPU speed is 1 GHz and a new virtual instance on new hardware would create a virtual CPU that would run at 2.8 GHz, then the server is likely a good candidate.

With these data elements, a solid assessment of the physical environment can be completed. Yielding in a list of potential candidates for virtualization. Next, cost data is needed to apply to the list of potential candidates. Whether chargeback models, rate cards, green dollar server cost, depreciation models or all of the above are used to calculate the cost of the physical servers, the cost of the list of potential candidates must be calculated. Also, the cost must be calculated for what the virtual infrastructure would cost and a solid guess must be made at how many virtual instances will go on each host. This should produce an average cost per virtual instance and can then be compared to the cost of the physical instances resulting in a potential save.

This data in a nice PowerPoint deck with fancy graphs is essential to selling the virtualization program to those that must sponsor and fund the effort.

The next entry will focus on the data and metrics needed post virtualization to prove the savings and remain efficient…