In this article, we will introduce to our readers some guidelines on how to keep virtual machines running efficiently. The key to this process of virtualized capacity planning is proper resource allocation. In this article, we'll walk you through best practices for capacity planning so that an enterprise's IT team can ensure that its virtualized infrastructure has enough resources to increase efficiency without wasting machine size. And given that improving hardware utilization is the fundamental purpose of server virtualization, it would be ironic (and embarrassing) for IT organizations to find wasted resources by failing to properly size virtual machines.

Properly sizing elements in a virtualized infrastructure is important, but not easy. By understanding workloads and classifying virtual machines according to real business needs, organizations can avoid wasting critical IT resources.

When you're looking at the size of a virtual machine and how its size affects capacity, a collection of questions comes to your mind. However, one most important point to always remember is that virtual machines are not free.

Given how easy it is to create virtual machines, people started thinking that virtual machines were free to use. This misconception is the source of over-allocation and sprawl of virtual machines; and the sooner you get rid of this misconception, the better chance your business has of a successful plan to properly size your infrastructure.

Classification, Monitoring and Baselining

One of the first steps in capacity planning is to understand your business workload. The types of workloads vary, ranging from databases (which consume a lot of memory) to transactional workloads (which are more CPU-intensive). While your organization is likely to have a variety of workloads, you should create different types of virtual machines to accommodate the various classes of workloads.

Without proper classification of virtual machines, it can be heart-wrenching for a system administrator or application owner to decide how big or small a virtual machine should be. The typical application owner or system administrator may not know or understand where their needs are. Unfortunately, when faced with this unknown, one naturally tends to think that the bigger the VM, the better. For them, proper categorization of virtual machines will not correct their mentality that virtual machines are free, but it will help them control and solve the problem.

The goal is not to create a different category for each workload, but to establish a set of categories that are appropriate for your enterprise's production operating environment. The number of categories an organization needs will depend on its specific business type and needs. Just remember that too few categories can easily lead to over- or under-provisioning of virtual machines, and too many categories are almost impossible to manage well.

As a general guideline, virtual machines tend to fit into a few natural categories: memory-starved, CPU-intensive, and storage-centric. The three options for creating small, medium and large virtual machines in these three categories allow for nine possible configurations to fit more virtual machines.

Matching category information to performance benchmarks can show you why virtual machines are placed in a particular category. When you have data showing that a workload can be run in a small to mid-tier tier without performance issues, your old beliefs about bigger VMs begin to change.

Appropriate virtual machine sizing and placement is only possible if you carefully monitor the content of your enterprise virtual environment and your next steps. If most of your virtual machines have two vCPUs or 6 GB or more of RAM, just create a baseline set of small virtual machines with one vCPU, 4 GB of RAM, and 20 GB of storage.

Predefined charts can display suggested placement locations for virtual machines. This is useful if your business happens to be running the exact same applications on hosts of the same size. Of course, in reality, applications and environments are unique on a business-by-business basis, so we need to set our own standards. The good news is that the job is fairly straightforward to accomplish.

Using a simple bell curve based on virtual machine benchmarks, your business can determine a middle range where most virtual machines will fall.

This helps you build a system of classes of small, medium, and large machines based on your unique environment, not just a collection of configurations from environments that may not fit your organization's specific business needs and challenges. Ideally, most virtual machines should be in the middle category. This allows your business to handle any potential growth.

challenge

The bell curve is very useful for the virtual machine category—as long as your business uses accurate data. What happens if the existing machine is not properly sized? If you apply a bell curve to existing data, you will create a set of categories that are not consistent with what you actually need. The resulting damage far outweighs the benefit.

Thankfully, many monitoring tools now seem to go beyond simple assignments. These tools measure the resources actually used by the virtual machine, not just the resources allocated. This sizing becomes the key to successful resource provisioning, helping organizations to clearly describe actual resource usage, and allowing your organization to create categories based on actual data.

Still, having a clear insight into the resources your business is using doesn't mean you should skip general housekeeping. By taking advantage of some best practices, you can prepare your virtual machines for an accurate baseline.

Storage scrubbing: Servers tend to collect large amounts of outdated data, often the fault of administrators. Typically, administrators copy the service packaging and installation files to the local server and usually do not delete them. These are usually in a dedicated folder on the C drive, or worse, on the desktop in the administrator's local server profile. Remember, deleting data simply moves it to the recycle bin, so be sure to clear each profile's data.

While only a few gigabytes of data per server may seem uncommon, the total amount of wasted space across multiple servers can be quite surprising. In addition to affecting the size of your enterprise virtual machines, these storage spaces can also affect backup performance and capacity.

CPU/Memory Cleanup: Virtual machines built from templates are generally resource efficient as long as the guest tools are installed and updated. This will be most evident during the migration from physical to virtual machines. When old machines are imported, they need to be cleaned up before being returned to production. Your business will need to uninstall vendor-specific hardware drivers and applications, as any service or driver that seeks out hardware that no longer exists has the potential to consume CPU or memory resources. While your business can find and uninstall vendor-specific items in the Programs section, don't forget to look at Windows Services and identify any hardware-specific services. Major features of services should be removed, including services that fail to start or that place notifications in the event log.

Monitoring: Management tools that come with enterprise virtualization environments often provide monitoring capabilities for metrics that are invisible to more traditional server monitoring tools. One of the keys here is its ability to monitor what resources are being used and what resources are being allocated.

As your business migrates more applications to virtual space, do you want to continue using these traditional non-VM monitoring tools? In fact, some environments may use multiple monitoring tools for the same server, as each department of the business Teams (networking, operations operations, server management, etc.) all have their own unique business needs. In fact, different tools are monitoring the same project through different means. With so many tools trying to do the same thing, it's even possible that an organization could inadvertently create a denial of service situation on its own server. Remember: monitoring is a good thing; over-monitoring is not a good thing.

Over-provisioning/thin-provisioning: Caps on memory and CPU resources, combined with thin-provisioning of disks, have good intentions, but the results don't always work. In fact, in many environments, overprovisioning masks the real problem.

This over-allocation tricks application and server owners into thinking they have more resources than they really are consuming. While some would say it's innocuous, what happens when they ask for another server? Application owners may continue to opt for a "bigger configuration" rather than asking for what is really needed.

If your business continues to support this type of overprovisioning, you end up with an overprovisioning environment where you don't know when your business needs additional resources. Classification can help reduce some excessive configuration requirements to a certain extent, and virtual administrators and application owners need to cooperate to stop these "little" lies.

If you find an overprovisioned VM——and you need to find it——you will need to adjust it back to the category settings you are creating. (Keep in mind that in most IT environments, removing resources from the application owner's virtual machine will likely incur considerable risk costs.)

All the graphs and charts you've prepared won't necessarily help convince application owners and administrators to let you delete resources, even if they can quickly add them back when needed. Therefore, this needs to be supported by the enterprise manager level.

To support your argument, base your bell curve on the resources being used and allocated. A good monitoring tool will be able to display this data. Armed with this data information, you can show how a collection of servers is classified in a broad category, even if these resources prove unnecessary in actual use.

As your business begins to size virtual machines, be sure to follow established baselines and monitor windows for these adjustments. Monitoring over a period of time does not provide enough data points to make precise adjustments. So your business will need to gather five to six weeks of performance data before making any adjustments. This allows you to cover a full month from start to finish, which also includes peak resource demands.

When evaluating the baseline for each virtual machine, be sure to exclude any spikes in backup or antivirus performance. These should be considered the exception, not the norm.

Baselines are the basis for possible changes to production VMs, so having a lot of rich data will be a big benefit.

proper IT

The reduction of resources can be somewhat personal to the application owner, so your business may need to take additional steps to ensure a smooth transition. This adjustment change is by far the most challenging in a scaled environment.

Reduction adjustments to CPU resources, drive size, or memory should only be made during downtime. While it may be tempting to update patches or the app itself, it's best to perform these tasks in a separate blackout window.

Otherwise, with a combination of adjustments, your business unit users will attribute any performance issues to reduced resources. This could have been due to software patches or upgrades hurting performance, but one would want to point to resource tuning. This perception will be difficult to overcome.

As you size your virtual environment by using baselines, classification categories, and cleanup, you have another critical task waiting for you: virtual machine sprawl. While cleanup is pretty good, be sure to keep an eye out for abandoned or too many virtual machines. The right size doesn't just refer to what's inside the virtual machine; it can also be the virtual machine itself.

Excessive and abandoned VMs are inevitable in a growing business organization with a constantly changing business, so this will be a problem. It needs to be solved constantly.

Capacity planning and virtual machine sizing is not a golden rule that applies to all virtual machines. Each enterprise's environment is uniquely determined by its specific business and its applications. However, the category framework will help you steer your environment to an optimized infrastructure by using precise monitoring data, reducing over-allocation and general cleanup.

The virtual environment gives enterprise organizations unlimited choices. Now is the time to tighten the reins a little and optimize the virtual machine configuration we already have. This means we are setting ourselves up for cost-effective, agile and successful IT operations operations.