Top 5 Tips for an Excellent End-User Experience (EUX) on Azure Virtual Desktop

September 28, 2021

With the Login VSI Summit coming up this week

I’ve been preparing content for my session “Sizing Azure Virtual Desktop and Selecting the Perfect Instance,” co-starring Pieter Wigleven (Twitter: @PieterWigleven) from the Microsoft AVD Product Team. I figured it would be good also to document what I’ll be covering in a blog as a reference you can use later.

I’ll review my Top 5 Tips for delivering a perfectly sized environment that provides a great end-user experience during my session. What’s more important is that it is not just a one-time thing… it is something you can use even after you have pushed your latest and greatest to production. It is based on the many experiences over the last year with some of our Fortune 500 customers who are moving quickly to Azure Virtual Desktop.

My Top 5 Tips for delivering an excellent EUX on AVD are as follows:

1. Know your users
2. Quickly find out what the performance counters are telling you
3. Get a real baseline to compare to
4. Find that sweet spot
5. Always be a hero

Let’s quickly cover these five points.

It’s all about the users

Login VSI is the industry standard for delivering a comprehensive workload that closely simulates the enterprise Task Worker, Knowledge Worker and Power Worker. Knowledge Workers are easily the most popular user type to test with. Today, with Login Enterprise, it has become easy to leverage our standard workloads and augment them with the line of business (LOB) applications that more closely represent what your users are regularly doing. This is important because your applications make your enterprise unique, and it makes a lot of sense to get some feedback on how those apps work in the context of a new, cloud-based, digital workspace.

The knowledge worker workload that I use to test AVD is based on my experience as a daily digital workspace user. I’m probably more of a Power Worker because I use many applications to do my job, and I’m constantly flipping between them.

Here is how my day looks:

• Browser – I start with my browser and catch up on current events while I drink my coffee
• Email – I jump right into my email as the morning tends to be the best uninterrupted time to catch up on my replies
• Research – During this phase, I’m researching data and information for my current projects. I mostly use Excel to look at information, as well as a web browser to research and learn, where YouTube tends to be a favorite
• Reporting – Using my research, I put together reports to distill the data into valuable information that can help my peers and the market
• Presenting – I do many webinars, so I build my slides in PowerPoint and often update them to fit a particular topic or audience daily
• Video and Chat Communications – I find that my usage of Teams and Zoom has increased almost 200% over the last couple of years. I use these platforms up to 5 and 6 times a day!

So, as we consider what type of workers we are trying to simulate, think of it this way…A Task Worker is a subset of a Knowledge Worker, and a Knowledge Worker is a subset of a Power Worker. Let’s take a closer look.

A Task Worker is someone who focuses on a specific task using specific applications. Typically, a Task Worker will use a browser and a LOB data application, like an ERP front-end or Excel.

A Knowledge Worker is probably most common and will use just about all Microsoft 365 applications, including Outlook, PowerPoint, Word, and Excel. They will use a browser like Edge or Chrome, and typically one or two LOB applications. They use all of these throughout the day. We are seeing web conferencing software utilization dramatically increase for this user.

A Power Worker
is a Knowledge Worker with higher utilization of compute-intensive applications, like development tools, CAD applications, and rich media.

The story performance counters can tell

A performance counter is simply information about the performance of a particular system. Think of the speedometer in your car that tells you how fast you are going. Basically, as we test a system, we collect this information throughout a test to see how it changes over time and provide some insight into what may be contributing to a degradation in the user experience.

There are many performance counters to choose from, but I have 7 that will tell me most of the story. They are:

• Application Start Times – how long does it take for the application to start – Login Enterprise
• User Login Times – how long does it take for a user to log into a machine/app – Login Enterprise
• User Input Delay – how long does it take an application to respond to input – Perfmon
• CPU and Memory Utilization – as we use more compute resources, a user’s experience will start to lag
• Disk Performance – I don’t care about IOPS, I care about how long it takes the disk to service a request, so I look at Seconds/Write and Seconds/Read
• EUX Score – This is a new technology that will be available in Login Enterprise this year, and you’ll hear more about it at the Login VSI Summit. I’m hooked on it. It gives me a simple score that tells me how good the user experience is. Does it remind me of the Windows Experience Index we used to get in Windows 7 and 8?

If you’d like to learn more, check out my last webinar on AVD Sizing.

Get a real baseline

This is new for me since it is now possible to use Login Enterprise Application Tests to evaluate a physical PC your users might be using. I’ve been leveraging the EUX score mentioned above in combination with other scores from Login Enterprise. It’s a great way to compare a new cloud-delivered workspace to those physical machines your users may have. Knowing that you can deliver an experience as good or even better than what they have today is a great way to get them excited about moving to a new workspace.

If you’d like to learn more, check out my last webinar on AVD Sizing.

Find that sweet spot

I generally expect 3 to 6 concurrent sessions per core from a session host as a rule of thumb. For the most part, I see this with AVD using Windows 10 Enterprise multi-session. You need to be careful, though, because a “core” is not the same as a “vCPU.” For example, an Azure D16s v4 instance has 16 vCPUs, but it is multithreaded from 8 cores. Most v3 and v4 instances work this way. So do a little math to figure out what you should expect. For example, I would expect a D16s v4 (8 cores) to deliver between 24 and 48 sessions, depending on the type of user.

If you’d like to learn more, check out my last webinar on AVD Sizing.

Always be the hero

OK. You’ve just pushed your latest tested configuration to production, and your users have never been happier. They all have access to the workspaces and applications they need, and they are performing great. You have all the tools set up to make sure it stays that way too.

Imagine a disk fills up and starts to slow things down, a cloud service provider has an issue, or an ISP link is slower than normal. How could you be the first to know? Easy, take the tests you used above, and place them into a continuous test cycle. Just one of your virtual users can be the first to alert you if performance degrades by just a little bit or if suddenly an app or workspace becomes unavailable. I’ve seen this all too often where the operations team discovers an outage and informs the right folks to fix the problem hours after the issue happens. Now you and your virtual user can be the hero when you keep downtime to an absolute minimum, or better yet, avoid any downtime altogether.

As always, thank you so much for your interest, and we look forward to hearing from you. Please stay safe and healthy and enjoy life…and request your trial today!