As the first wave of server virtualization
If you can take the data collected during virtualization health checks and identify areas where your customers can improve on their resource utilization, you can save them additional money and make these projects even more profitable for your business.
You may think that a virtualization project is a success just because it delayed the purchase of a dozen or so servers But by adding a resource utilization project, you can make the deployment even more successful.
The underutilization of virtualization projects begins at the servers themselves. Typically, customers go out and buy new, more powerful servers to host the virtual infrastructure. Herein lies the problem: Those new servers are so powerful, and the virtualization software is so efficient, that when your customers virtualize their initial sets of servers, it does not stress resources. In fact, in many cases resource utilization improves. Also, most customers only virtualized the low-utilization, low-risk servers first, which are easy for the new virtual hosts to handle.
Assess workloads before resource utilization projects begin
Now, especially because of the economic downturn, your customers are looking for ways to further improve their resource utilization. Before exploring resource utilization projects with them, you need to expand the scope of your assessment to analyze servers that are not currently virtualized.
Examine the I/O workloads of these servers and the applications that reside on them. You should also capture information about these servers to determine their potential for becoming hosts in virtual clusters. It is important that no server be ruled out because it is too I/O intensive. Today's technology lets you virtualize an increasing number of workloads that just last year would have been ruled out.
For example, you can use raw-device mapping on a workload to make sure it has exclusive use of specific storage I/O bandwidth. Brocade has added Fibre Channel quality of service (QoS) capabilities to storage host bus adapter (HBA) cards, which allows for prioritization of the storage I/O within specific virtual machines by allocating additional buffer credits to higher-priority workloads. And on the IP network, Neterion Technologies, SolarFlare and other companies are offering cards that allow for prioritization of IP traffic to particular virtual machines.
Technologies like these will become more commonplace because they address the two typical concerns of critical applications: service levels and their effects on resource utilization. Service levels are able to be maintained because these technologies prioritize the workload's I/O traffic and they minimize the effects on resource utilization by isolating I/O traffic.
The next step is to do a gap analysis on the workloads that have the potential to be virtualized. What are their requirements as far as uptime, network I/O and storage I/O? And what are the available resources of the virtual hosts? The basic goal here is to make sure the guest operating systems on a particular host are balanced. If you have 10 virtual machines, make sure not all of them are demanding storage I/O applications. Also, consider that some applications are resource-intensive at particular times of the day, week or month.
For example, you could have a virtual workload made up of a payroll application that is very I/O-intensive twice a month grouped with an application that is I/O-intensive on the weekends. You could even create a plan as to when these virtual machines should be moved to another host -- or even unvirtualized, as discussed in part two of this series -- for specific times of the year.
Server repurposing in resource utilization projects
As part of the optimization assessment, you should account for viable migration targets. One of the values of virtualization is the ability to move workloads from host to host. Mapping out a prioritized list of migration targets for your customers also helps them, because knowing where they should migrate a virtual machine to is far more valuable to them than just guessing.
The physical servers that used to house these workloads are viable candidates to act as virtualization hosts because they were originally running more demanding applications. This is a good time to introduce server repurposing solutions to your customers. Egenera, Scalent and other vendors offer software that allows you to leave these servers in a powered off state and then, when additional processing power is needed, to power them on, assign them to boot a virtual host image and then transition workloads to them.
Server repurposing gives you and your customers a safety net as you try to maximize capacity. If an unforeseen event occurs that stretches the limits of the infrastructure too far, you can simply point and click to start up spare processing power. What makes these solutions so unique is that the systems can remain in a powered off state consuming no power until additional processing resources are needed. Also, the use of servers under the control of a server repurposing application does not need to be predetermined; they are determined at boot time.
You can use a spreadsheet to track the information needed to make these assessments. But in today's dynamic environment, specific tools from Tek-Tools, Virtual Instruments, VKernal or other companies may have greater value and produce more accurate results. Some of these tools can actually simulate the effects of virtualization of a particular workload, so you can predict its resource utilization.
Even with these tools, the human element cannot be factored out. Some do not account for the advanced virtualization capabilities provided by hardware vendors. Your knowledge and expertise in these environments, plus your understanding of virtualization-aware hardware, is invaluable.
By combining the tools' findings with what you know the technology is capable of, you can push customers' virtualized environments further, increase the success of the overall virtualization projects and help the customers see more return on their investments in your assessment services.
This was first published in March 2009