Virtualization technology has matured and is a critical infrastructure component in most modern data centers, especially as a trusted part of production environments. As our use of, and reliance on, virtualization technology continues to grow, it is important that we manage the technology using best practices. Here are five best practices to help ensure success in leveraging virtualization technology:
By Reza Koohrangpour
1. Monitor the Entire Infrastructure Stack
Virtualization touches almost all aspects of the data center including servers, storage and networking. While most virtualization vendors provide tools to monitor and manage their technology, they rarely give an IT organization a comprehensive view into the entire stack of technology that virtualization touches. For example, understanding network traffic at a physical port, or being alerted to drive failures on a SAN, might be important to the performance of your virtualized environment. These are a just a few examples of things that impact the full virtualization stack but are typically outside the purview of dedicated virtualization monitoring tools. Ensure your tooling provides you visibility into all parts of your infrastructure stack.
Related: Where Next for Virtualization?
2. Monitor Inside Virtual Machines
For organizations trying to maximize utilization it can be valuable to monitor virtual machine (VM) usage from the perspective of the hypervisor, as well as the operating system inside the VM. The way CPU and memory resources are scheduled to a VM from the outside may not match what the demand for resources looks like from the inside.
Understanding the performance of a VM in terms of the physical resources it is consuming is critical. However, it is just as critical to monitor what is happening inside the VM itself. In particular, monitoring the applications inside the VM is the only way to really understand if everything is running as expected. To provide a good end-user experience you need to understand when critical services have stopped or a database can’t keep up with the volume of data being read and written. Remember that the VM being healthy is a means to an end; it’s more important to know if services and applications are being delivered.
3. Use the Right Metrics
Hypervisor vendors have learned they need to do a good job helping users figure how to measure performance in the right dimensions. To that end, they publish papers that help us identify the key metrics that will point to performance issues. Many of the metrics are ones we have used for years – CPU usage, memory usage, disk IOPS, network IOPS, etc. Watching these standard metrics on both physical servers and VMs is a must.
There are also metrics that can help us hone in on issues specifically related to virtualization. VMware, in particular, does a good job of surfacing metrics that point to these types of issues, such as CPU Ready, Memory Ballooning, and CPU Co-stop. For example, CPU Ready indicates whether a physical host is able to keep up with the CPU demand from the virtual machines running on it. Include these metrics in your monitoring strategy as well.
4. Keep an Eye on the Future
Virtualization has greatly increased our agility in the data center. In many environments, the time it takes to provide requested resources to satisfy project demand has decreased from months to minutes. A side effect of this agility is that we are running many more workloads (operating systems) today than we ever have before. On the positive side, companies can quickly execute on new projects. On the negative side, companies are consuming more and more compute, storage and networking resources. Today’s IT organizations are being asked to run as lean and agile as possible while still providing enough resources to execute on future demands. Purchasing too much resources results in waste and contradicts the desire to be agile, while supplying too few resources means either a decrease in the ability to execute on new projects, or an IT environment that is potentially less reliable or stable.
To the extent that past trends can help predict the future, they can be used to help guide a company’s hardware requirements for the future. There are four primary metrics that companies should be planning around – CPU utilization, memory utilization, disk utilization, and disk IOPS. By projecting the past usage of these key metrics against the capacity available for each company, you can ballpark when they will need to purchase more hardware to satisfy future demand. Having a detailed understanding of utilization versus capacity can also help an organization keep these resources in balance.
There is another related side effect of virtualization, and that is waste. IT organizations need ways to track this waste so they can reclaim resources when they are no longer being used. For IT organizations operating as private clouds, a second technique to control waste should be evaluated: oversubscription. While most of us don’t appreciate the oversubscription practices of the airlines, car rental companies or even our doctors, they do help those organizations drive down costs.
5. Take Advantage of All the Virtualization Technology
Virtualization technology has evolved beyond just the hypervisor. Most virtualization vendors are building software-defined storage and software-defined network technologies. These technologies create a layer of abstraction similar to what the hypervisor provides from the server. These abstractions help improve the management, utilization and the cost effectiveness of storage and networking. The investments made in learning, understanding and using these new technologies will almost certainly provide benefits in the long run.
In addition to technologies that abstract hardware from applications, virtualization vendors have created new technologies that can make the data center much more efficient and reliable. One of the key challenges with virtualization stems from multiple VMs running on a single physical system. Virtualization vendors provide technology to recover VMs on other hosts if the current hosts fails, and automatically balance resources, including powering off unused resources. There are many other ancillary technologies that can complement virtualization, from new improved approaches to managing disaster recovery and backups, to running and managing desktops.
Virtualization revolutionized the data center. Data centers today are more agile, cost effective and better utilized. By focusing on these five best practices we can leverage virtualization to its full potential.