Virtualization sprawl occurs when the amount of VMs on a network exceeds the amount the data center administrator...
can realistically manage. The beauty of virtualization is that it enables users to quickly and easily spin up new virtual resources.
However, when left unchecked, these underused and unneeded resources can put a serious strain on computing resources and affect data center performance and efficiency. Therefore, it's in an organization's best interests to mitigate the effects of virtualization sprawl or, if possible, prevent it before it even happens.
Organizing VMs so that they consume resources evenly, assigning a cost value to virtualized resources and limiting administrative access to the data center are all key elements of an effective VM sprawl management plan.
Organize VMs to avoid stranding capacity
Every VM requires computing resources in order to function. But even though allocated resources are virtualized, those resources still translate to physical resources on a host server. A host server only supplies a finite amount of CPU, memory and other resources, so there's always a practical constraint on the number of VMs that can reside on any given system.
Ideally, VMs would consume the host's resources evenly, but this is rarely ever the case. Practical VMs often utilize more memory or more CPU than others -- meaning that a host server typically runs out of one resource, leaving other resources unusable. Those unusable resources are called stranded capacity.
For example, if a server is populated with memory-hungry VMs, that server might run out of memory before running out of CPU cores, leaving those leftover CPU cores unusable. This means that admins can't mount additional VMs because there is inadequate memory. In this example, the leftover CPU cores would constitute stranded capacity on that particular host server.
Stranded capacity is basically a waste of capital because the business would then need to purchase additional servers -- even though some resources might remain unused on each server. Administrators can use tools, such as the performance charts, alerts and logs in VMware vSphere, to spot stranded capacity on each server, and then migrate VM instances to even out resource use. For example, admins could deploy VMs with a mix of CPU and memory use so the maximum amount of a server's resources is utilized.
Even though VMs aren't tied to specific servers, the way in which VMs are organized and deployed on physical servers can substantially affect server utilization and cost efficiency. This practice also helps to optimize server consolidation and improve VM sprawl management.
Employ chargeback/showback methodologies
One of the central benefits of virtualization technology is its capacity for speed and flexibility. Where a traditional physical server has to be deliberately justified, purchased, deployed, configured and maintained, a virtual instance can be created -- and destroyed -- with no material costs with just a few mouse clicks. This means a VM can run for years, but it might also be needed for just a few weeks, or even a few hours.
The great challenge for virtualization is that such speed and flexibility often conflict with increasing sensitivity to regulatory issues and business governance. VM owners are often unwilling to release unused virtual instances for fear that the instances might be needed later, or might potentially expose themselves or the business to other risks. VM owners instead opt to hang onto virtual instances indefinitely, regardless of whether those instances are actually used, resulting in virtualization sprawl.
Proper tools and provisioning prevent sprawl
In part one of this two-part series on virtualization sprawl, we take a look at how a virtualization administrator can put an end to sprawl by using the right management tools to implement policies and right-size provisioning to prevent resource waste.
One means of enforcing VM sprawl management is to assign a cost to virtualized resources, and then charge those costs against the VM owner's monthly budget -- a practice called chargeback. With a chargeback model, business departments and groups effectively pay for every VM instance that they request and keep. Thus, business leaders are acutely aware of the virtualized resources they consume, and they have a financial incentive to review those assets periodically and make pragmatic, proactive decisions about which instances to keep or dismiss.
An alternative to chargeback is the showback model, where resource consumption is also tracked and reported back to business departments and groups. As with chargeback, showback reporting might also contain valuable information about the virtual resources being utilized, but those costs aren't recovered from the VM consumer. Still, the regular summary of resource usage can be enough to promote resource conservation, and it is critical to VM sprawl management.
Limit the authority to create virtual instances
Finally, virtualization sprawl can be mitigated by limiting the number of administrators or other staff capable of creating virtual instances in the first place.
This restriction is anathema to modern precepts of cloud computing founded on user self-service and commonly relying on chargeback. However, establishing such a bottleneck in a virtualized environment and deploying suitable tools can enhance the administrator's ability to oversee resource use and provide greater familiarity with virtual resource usage that helps identify heavy usage and potential candidates for removal.
Control, visibility and cost
Ultimately, an ideal VM sprawl management plan should consist of a combination of control, visibility and cost awareness. Policies are central to controlling virtualization usage in accordance with business goals. Tools offer the nuts and bolts needed to identify resources and show how those resources are actually being used -- or not. And implementing cost awareness carries the financial incentive for leaders to watch resource usage and release unimportant virtual instances.