Virtual desktop infrastructure is highly sensitive to poor performance, perhaps even more than any other type of...
compute environment. Poor performance can affect many users at once and make life difficult for administrators. In this article, we'll take a critical look at the elements that make up a virtual desktop infrastructure and discuss which areas would benefit from potential improvement and optimization.
The best way to look at a virtual desktop infrastructure (VDI) environment and to stay on top of VDI performance monitoring is to split it into three separate areas: hosts, guests and networks.
Creating a reliable host
It goes without saying that hosts should be in a good state of repair. An optimized, fully functioning server is a good host. You can use the inbuilt vSphere performance monitoring at the host level to see how the hosts are performing in general and how they can be used to triage any potential issues on the horizon.
Knowing how your environment ebbs and flows is important. You will find that you have peaks and troughs throughout a specified period -- usually at the start and end of a shift. Having a known baseline makes VDI performance monitoring easier because it shows when items are out of their normal performance profile.
Guests lead to better VDI performance monitoring
Manage the experience
When monitoring guest health, the key metrics to keep an eye on are the CPU ready time and whether any memory swapping or ballooning is going on within the environment. CPU ready time is the amount of time a virtual guest has to wait for access to the CPU -- for example, to perform instructions. If virtual guests are ballooning or even worse, using swap, it means that you have some serious memory contention going on.
In an environment that has either of these items, you need to add RAM and/or CPU resource at the host level. Your users won't be happy if these values are not at or near zero.
Know your estate performance
VM performance monitoring is critical within VDI environments. Monitoring allows you to have a proven baseline for performance. It removes people's perceptions and replaces them with facts.
VM monitoring allows administrators to drill into the machine and see what is happening inside, saving admins time and guesswork spent educating users about the cause behind a slow VM. Tools like vRealize Operations and Liquidware are ideal for these scenarios.
Build for VDI, not for desktop
Many VDI environments, particularly smaller ones, use builds designed for desktops and laptops. VDI guests are designed and implemented differently.
If you aren't already using linked clones, you're missing out. With linked clones, the whole system becomes one master disk and the machine's specifics are stored in delta disks.
When building your Windows image -- or even VM if you aren't interested in using linked clones -- don't waste compute cycles and memory resource. By turning off or otherwise disabling certain items within your Windows master image, you can conserve resources. These items include:
- Turning off the Windows search service;
- Reducing the VSS snapshot -- VDI environments rarely use these;
- Reviewing the Windows services and disabling ones that are not in use;
- Reviewing desktop customization settings, such as graphic effects, and turning them off to avoid resource consumption;
- Not using screensavers -- if you have 500 VDIs and each one uses 100 MHz of CPU for screensavers, that is 5 GHz of licensed CPU power that is being used for no good purpose; and
- Optimize antivirus and backup software for VDI; loading a copy on each VDI is wasteful and inefficient and several vendors now have VDI-optimized AV scanning.
Leverage built-in VDI functionality -- and possibly save money
One issue I frequently see when people use linked clones is that whoever builds the image leaves applications off the list that users need. It doesn't take long for several developers or users to add additional tools and disk space consumption shoots through the roof.
An administrator can simultaneously increase performance and reduce cost and usage with linked clones, provided they do so properly within a good infrastructure. Just make sure you have an appropriate set of packages and the ability to add more.
The clever administrators may use VMware ThinApp to add and remove packages from a build. This reduces the issue described above, but in order to use it, you need to use a known good build. ThinApp is appropriate for some situations, but not all.
Using linked clones can significantly decrease the management overhead and consumed resources. For example, a professional client I recently consulted for was using a nonlinked clone setup and was patching its desktop environment.
Due to the huge number of desktops, the patches were normally applied to groups of machines. A new administrator applied a patch to the entire VDI environment and effectively brought the system to its knees in terms of performance due to hundreds of desktops trying to update at once. Had the company in question used a linked clone setup, updates would have happened in a quicker, more efficient manner since you don't need to deploy updates to each individual VM with linked clones. Linked clones update the replica, and then the machines are updated and rebuilt from the replica.
Assessing network performance
When all is said and done, network performance is also an important factor. I often hear users complain that their sessions are slow, but many of them were coming to the U.K. from Australia, so a change in performance consistency is to be expected. Unfortunately, not even I can defy the laws of physics with regards to latency. However, I can reduce the effects.
Obviously, moving the VDI farm isn't really a feasible idea, but you can reduce the performance bottleneck by using PC over IP (PCoIP) rather than Remote Desktop Protocol. RDP is an old protocol that doesn't deal with latency very well; PCoIP was built with that in mind. If you are not using PCoIP, you are missing a sizeable performance boost.
I hope that these items have given you some overview as to what can and often does drag down VDI performance. Not all the items may apply, but if nothing else, you should conduct regular VDI performance monitoring.
Check out the top VDI monitoring tools
Steer clear of VDI performance issues with monitoring tools
Unified tools to monitor VDI and your whole environment