Nutanix and AMD Break the User Density Bottleneck

The focus shifts to meeting business requirements, where it should be

By Kees Baggerman

April 12, 2021 | min

When provisioning End User Computing (EUC) solutions (providing virtual desktops and applications to end users) hardware and software limitations have kept the conversations centered on finding a balance between maximum density and the most optimal user experience. In this blog, I’ll describe how innovation from Nutanix and Advanced Micro Devices, Inc. (AMD) effectively eliminate this constraint, allowing organizations to focus on fulfilling business requirements and justifications instead.

End User Computing typically has four important metrics:

  • Disk latency
  • Disk capacity
  • RAM
  • CPU

Before we can even start talking about the additional complexity of the upper stack, we must address a few key issues:

Reduce Complexity

Ever since its introduction, Nutanix has been able to reduce complexity with a focus on both improving performance and lowering the administrative overhead. 

When desktop virtualisation (VDI) was introduced, the industry saw an increased demand in lowered disk latency and higher disk capacity. A number of solutions were introduced into the market, varying from built-in caching of reads by brokers (MCSIO, PVS RAM Cache with Overflow to disk, VMware® VSA/CBRC) to storage systems including much faster disk setups, such as All Flash and later on NVMe-driven configurations.

System RAM has always been relatively straightforward to specify: look at the number of VMs you want to run on a system and multiply that by the RAM per VM and you’ve got your User Virtual Machines (UVM) RAM demands. 

CPU traditionally is the most capricious metric because there are so many elements in sizing CPU that you have to account for, such as the operating system, physical to virtual CPU ratios, VM resource sizing, applications, end users, and processes. The list goes on, making accurate calculations for CPU utilization a cumbersome task. 

In comes innovation:

The Nutanix® Distributed Storage Fabric (DSF) has taken away the complexity around disk latency and capacity. Our storage fabric provides low latency access to the storage layer, delivering excellent user experience for your virtual desktops and applications. 

RAM configuration for Nutanix software is similar to the process above -add CVM memory overhead and off you go. The Nutanix configurations can provide up to 3072GB per node, ensuring that RAM won’t be a bottleneck anymore. 

CPU has been more difficult to size for. Up until recently, discussions usually centered on the maximum density of a system with minimal impact to user experience. Innovation from chip manufacturers is changing this conversation. The Spectre and Meltdown vulnerabilities spurred a surge in innovation from Intel, resulting in Cascade Lake and the Cascade Lake Refresh series. This enabled higher core counts and clock speeds, both vital considerations for CPU sizing. But maybe even more importantly, Intel’s overall innovation drove AMD to look at the server CPU market and revamp their interest. Nutanix has been partnering with AMD to deliver a new offering based on the AMD EPYC CPU series. Based on this partnership, the Nutanix Solutions Engineering team for EUC has been collectively working with AMD to test these new CPUs for density and user experience and the results have been very positive. 

We received the following configuration from AMD to test using Login VSI as our standard test tool for EUC testing. 

Node Type CPU RAM Disk
HPE DX385 2x AMD EPYC 7542 (2.9GHz) 1024GB 2x SSD, 10x HDD

The results of our single node tests are very promising and we see consistent and linear results scaling up the number of nodes; we had to scale down RAM configuration per VM to see where the actual limitations of this CPU series were. Our tests were conducted using Citrix® Virtual Apps and Desktops 1912 Update 1 as a broker. On the VDA side we used Windows 10 1904 with 2 vCPUs and 2.5GB. As previously mentioned, we lowered the RAM from the initial 4GB to 2.5GB to see what the CPUs were able to run in terms of number of desktops:

Test # of VMs Workload AOS Hypervisor Deployment VSIbase VSImax
1 360 Knowledge worker 5.19 AHV MCS 758 336
2 360 Knowledge worker 5.19 AHV PVS 732 337

This provided some additional insights and we decided to run the same test with a higher vCPU configuration, moving from the 2 vCPU configuration in our previous tests to 3 vCPUs per desktop and testing at 80% host CPU utilization, which is a reasonable host utilization:

Test # of VMs Workload AOS Hypervisor Deployment VSIbase VSImax
1 235 Knowledge worker 5.19 AHV MCS 711 Not reached
2 235 Knowledge worker 5.19 AHV PVS 714 Not reached

For clarification of the results, a VSIbase score under 1000 means user perceived user experience is within tolerance, and the lower score the better. VSImax indicates the maximum number of users per node. In the second round of testing, we achieved better user performance and no perceived bottleneck based on an individual node. For more information on interpretation of LoginVSI results, see this document.

What our testing showed is that with these newer CPU sets we can achieve incredibly high densities and corresponding lowered application response times, even to a point where the discussion changes from, “What will be the maximum density of this system?” to “What will be my maximum failure domain--that is, if we lose a node, how many desktops do I want to have affected?”

In my opinion that’s right where we want to be. When designing End User Computing infrastructures, the preeminent concerns should be business requirements and justifications, not hardware and software limitations. And that’s how innovation is changing discussions.

If you are interested in learning more about the AMD nodes currently offered by HPE, here is a page for more information and more information on EUC is available on our main solutions page.

Break through your EUC density bottleneck today with capacity and solution planning by Nutanix Sizer.

© 2021 Nutanix, Inc.  All rights reserved. Nutanix, the Nutanix logo and all Nutanix product, feature and service names mentioned herein are registered trademarks or trademarks of Nutanix, Inc. in the United States and other countries. Other brand names mentioned herein are for identification purposes only and may be the trademarks of their respective holder(s). This post may contain links to external websites that are not part of Nutanix does not control these sites and disclaims all responsibility for the content or accuracy of any external site. Our decision to link to an external site should not be considered an endorsement of any content on such a site. This post may contain express and implied forward-looking statements, which are not historical facts and are instead based on our current expectations, estimates and beliefs. The accuracy of such statements involves risks and uncertainties and depends upon future events, including those that may be beyond our control, and actual results may differ materially and adversely from those anticipated or implied by such statements. Any forward-looking statements included herein speak only as of the date hereof and, except as required by law, we assume no obligation to update or otherwise revise any of such forward-looking statements to reflect subsequent events or circumstances.