Virtualization sprawl -- also known as VM sprawl -- can be a real headache for data center administrators and a...
strain on enterprise budgets, but it doesn't have to happen. Let's look at how you can control VM sprawl in the enterprise with management tools and right-size resource provisioning.
According to Parkinson's law, the demand for a resource eventually expands to match the supply of the resource, if the price is zero. In IT, this principle applies to resources, such as storage, where factors like burgeoning file sizes and extended retention choices eventually drive up utilization until storage capacity is exhausted.
Parkinson's law also rears its ugly head in virtualized data centers in which VMs and storage instances can be easily requested and quickly provisioned with common tools, including VMware vSphere. Unlike traditional physical resources, which require laborious budgeting, justification and deployment, the speed and ease of virtualization enable users to provision resources in a matter of minutes without capital investment.
The trouble is, once created, users are often reticent to discard unneeded resources just in case they're needed in the future -- regardless of whether those resources currently do any meaningful work for the business. At the same time, IT organizations might lack the governance policies and tools necessary to adequately track and manage virtualized resources.
This confluence of factors can potentially enable unused VM and storage instances to be ignored and forgotten, wasting and even losing valuable resources within the data center infrastructure.
There are a number of ways for a virtualization administrator to mitigate the effects of VM sprawl. In part two of this series on sprawl prevention, we'll take a look at how practices like proper VM organization and implementing chargeback/showback can save resources.
The net result is that virtual instances are created faster than they are removed, which leads to virtualization sprawl that eventually exhausts computing resources and forces unwanted spending on new resources without adding any new benefits or capabilities to the business. The great irony of virtualization sprawl is that it threatens the consolidation benefits that make virtualization attractive and cost-effective in the first place.
Manage the provisioned resources
Management can provide insight and exert direct control over the virtualized environment, and that is the single most vital method to control VM sprawl. Management generally comes in two forms: policies and tools.
Policies guide the creation and retention of VMs and storage instances. Policies can -- and should -- cover a wide range of topics, such as defining who is authorized to request virtualized instances, the uses of those instances, the period of time for which those resources should be retained before review and recovery, how those instances should be backed up and protected, and so on. Policies are critical to define the lifecycle of virtualized resources within the business, and they can play a pivotal role in maintaining good business governance and regulatory compliance through copious control of the IT environment.
Tools like VMware vCenter Server and VMware vRealize Suite are the means of codifying and implementing these policies. Management tools can summarize free and consumed resources and display all of the hosts and virtual instances within the business.
Basic tools can create and remove virtual instances, such as VMs and storage; migrate VMs between physical host servers; and perform simple reporting on operational status and performance. These tools can also implement lifecycle management features, such as identifying VM instance owners, reporting on recent usage or how long the resources have been idle, and so on.
More advanced tools can also spot interdependencies, resource shortages, help with troubleshooting within the virtual environment and even provide a platform for constructing a private cloud.
These tools can also include automation and orchestration capabilities that speed resource provisioning, standardize provisioning through configurable resource templates, queue unused or expired instances for shutdown or removal, and orchestrate any necessary approvals for new instance provisioning or old instance removal.
This combination of features enables IT administrators to quickly see all of the virtualized instances and identify abandoned or unused instances for removal -- key elements in the fight to control sprawl. VMware already provides such features in vSphere with Operations Management.
Practice right-size resource provisioning
The resource demands of a virtual instance can sometimes vary, and IT administrators frequently engage in the practice of overprovisioning VM and storage instances to ensure adequate resources and preserve performance for the workload. Overprovisioning is typically performed pre-emptively before the workload actually needs it.
The problem with overprovisioning is that it wastes valuable virtual resources. If the virtual instance works normally, it doesn't need the additional resources. Those additional resources are also lost to the business if the virtual instance is unused or abandoned. When sprawl is rampant in a data center, overprovisioning resources leads to more resource waste.
IT administrators must perform due diligence to understand the needs of the workload and to allocate the appropriate amount of resources to it from the start to control VM sprawl. Management tools may provide a selection of templates for standardized instance deployments -- each template offering a blend of CPU, memory and storage for specific workload types -- so select the template that is best-suited for the particular workload.
Workloads with highly variable or unpredictable resource demands or those that require higher levels of availability might benefit from advanced deployment approaches, such as a load-balanced VM cluster, perhaps with an auto scaling capability. For example, VMware partner KEMP Technologies offers Virtual LoadMaster as a virtual load balancer for VMware VMs.