Servers: The Right Compute
Showing results for 
Search instead for 
Did you mean: 

10% improvement in scaling with VDI-in-a-Box 5.2 and HP ProLiant DL380p


Guest Blog written by: Frank Anderson, Citrix Systems


Back in August of 2012, when it was hot and humid, and the sun never seemed to set in South Florida, I performed extensive testing to validate specific HP server and Citrix® VDI-in-a-Box™ configurations. Using HP ProLiant DL380p Gen8 servers, I tested Citrix VDI-in-a-Box on Microsoft Server 2008 R2 with Hyper-V to come up with a set of sizing and scalability guidelines for pooled and personalized desktops. I blogged about the reference architecture and my test results, which can be found here.


ViaB 5.2 image with DL380.jpgAs the days of summer came to an end, the temperatures started to drop and the days got shorter,, Before I knew it, it was December and Citrix released version 5.2 of VDI-in-a-Box—the first Citrix desktop virtualization product to be supported and validated on the updated version of Microsoft Server 2012 with Hyper-V. Naturally, I was anxious to see how VDI-in-a-Box 5.2 would scale on this new hypervisor release but I was happy to find that when I ran the same tests using the same hardware configuration, the user density improved on Hyper-V 2012—as much as 10%! During this testing, I also wanted to see how Citrix VDI-in-a-Box would scale across a two-server grid. In an additional dual-server test with Hyper-V 2012, I demonstrated linear scalability. I’ve updated the white paper “HP Client Virtualization SMB Reference Architecture for Citrix VDI-in-a-Box” to add these results, summarized below.





Summary of Results

In earlier testing with Hyper-V 2008 R2 and VDI-in-a-Box 5.1, I was able to run 150 desktops (either 100% pooled or a 70/30% mix of personalized/pooled desktops) without reaching Login VSImax, which indicates acceptable user experience at that capacity. In my Hyper-V 2012 testing with VDI-in-a-Box 5.2, I was able to scale to higher user densities: 165 users for pooled desktops, 160 users for a 70%/30% mix of pooled/personal desktops, and 330 pooled desktops on two HP ProLiant servers. Again, Login VSImax was not reached in any of the tests.


Single server configuration

Pooled desktops

70/30% Mix

VDI-in-a-Box 5.1



VDI-in-a-Box 5.2




Test Description

Just to review, the reference architecture defines a pre-sized and easy-to-implement configuration to deploy as a scalable VDI building block. The configuration is based on HP ProLiant DL380p Gen8 servers, the Citrix VDI-in-a-Box software, and the Microsoft hypervisor.

I tested pooled desktops along with a mix of 70% pooled and 30% personalized desktops. (Pooled desktops use a single, centralized virtual desktop image, while personalized desktops use both a published image and a customized personal vDisk in which users can install their own applications and data.) To simulate a real-world scenario, I set up the required Login VSI applications: Microsoft Office 2010, MindMap, and Adobe Acrobat Reader in the personal vDisk, while I installed Flash, Shockwave, Java, and the Bullzip print driver in the base image (since these are typically foundational applications).



For each test run, I followed this sequence of steps:


1) I started a script that invoked PerfMon scripts to capture comprehensive system performance metrics, and then initiated the workload simulation portion of the test in which Login VSI launches desktop sessions at 30-second intervals.


2) Once the desktops are logged in, the steady state portion of the test begins. During steady state, Login VSI tracks application performance statistics, looping through specific operations and measuring response times at regular intervals. Response times are used to determine Login VSIMax, the maximum number of users the test environment can support before performance degrades consistently. 


3) After a specified amount of elapsed steady state time, Login VSI logs off the desktop sessions. After all sessions are logged off, I stop the performance monitoring scripts.


4) Lastly, I process the Login VSI logs using VSI Analyzer VSI Analyzer and PAL (the PerfMon CSV tool) to produce the graphs and metrics that follow.


The graphs below show the results for a workload of pooled/personal desktops (70/30% mix) on a single server. As shown in the Login VSIMax graph, the configuration achieved a user density of 160 users without achieving Login VSIMax. The Logical Processor Run-Time graph illustrates processor utilization of the Hyper-V server during the test. For complete test results and metrics, see the full report.


Login VSIMax


Login VSIMax


  Logical Processor Run-Time


 Logical Processor Run-Time



The goal of my testing was to see how the new 5.2 release of VDI-in-a-Box performed in conjunction with Hyper-V 2012. A 10% improvement in density for pooled desktops was a pretty exciting result - well worth spending long days in the test lab!


But I was also curious about the new release of Microsoft Office 2013 and how it would compare to previously tested workloads based on Microsoft Office 2010, so I tested that as well. The results might surprise you, so I promise to blog about them in the near future. (Hint: if you can’t wait until then, see the reference architecture paper.


Of course, there are always more test scenarios to explore than there is daylight -  no matter what the season but I’ll do my best to bring you more results.  Until then, if you’d like more information on VDI, please click on the links below.


For more information:



About the Author



In HPs Technical White Paper "HP Client Virtualization SMB Reference Architecture for Citrix VDI-in-a-Box" underneath page 5 there is the following statement:

In this solution, SSDs store the “golden” desktop image and help to significantly increase desktop performance, enabling higher user densities.

When we go to the Citrix VDI-in-a-Box Reverse Server Sizing Calculator in fill out the components we receive a excel spreadsheet which under "Oure Recommendations" says:

Storage configured for Golden Images
"If you only have a single datastore, all components are stored here - golden images desktop instances and personal vDisks. 15K SAS drives are recommended for the storage of golden images.

So who is right?
Kind regards,
28-30 November
Madrid, Spain
Discover 2017 Madrid
Join us for Hewlett Packard Enterprise Discover 2017 Madrid, taking place 28-30 November at the Feria de Madrid Convention Center
Read more
HPE at Worldwide IT Conferences and Events -  2017
Learn about IT conferences and events  where Hewlett Packard Enterprise has a presence
Read more
View all