This post was originally published on this siteCustomers and IT professionals I encounter are increasing interested in and challenged by the emergence of convergence as a new model ...
Comparing 2000 Active VDI Desktops for XenDesktop MCS and PVS on FlashStack CI
When moving forward with a Citrix XenDesktop deployment, one of the most important early decisions is figuring out which deployment method makes the most sense for your scope of operations, users and applications. There are advantages and challenges for both Machine Creation Services (MCS) and Provisioning Server (PVS); so it can be a tough decision that will likely have long-lasting implications on your VDI project. We decided it would be interesting and hopefully helpful to compare and contrast how those two solutions perform on FlashStack Converged Infrastructure at a reasonably high scale in order to show what you might expect for each solution by using Login VSI as a workload generator.
In Figure 1 we can see a connectivity diagram for our FlashStack test environment. Notice that it is fully redundant with no single point of failure.
In order to minimize variables and make this as much of a direct comparison between MCS and PVS as possible, we took the following steps:
For each solution, we will highlight the following metrics:
We found that there was no major differences between the MCS and PVS technologies on any component of the FlashStack – even when running 2000 concurrent desktops. ESXi CPU and memory utilization was close to identical in both solutions with sufficient headroom for additional operations to take place.
Figure 2 shows an overlay of the two twenty host Login VSI VSImaxv4 charts (read more here about this metric) and we can see that in this case MCS (purple) provided a slight performance advantage over PVS (blue) but it is not major enough to warrant making a design decision on this factor alone.
One of the more important measures of performance for VDI is the amount of time it takes from when a user clicks ‘connect’ in Citrix Receiver to when they have a usable desktop. Below we can see the logon times for each virtualization technology we again see near identical performance – though with PVS we did encounter a bit more variability than MCS in all tests.
Finally, when we compared array performance (the below measurements were taken when the 2000th desktop launched in each simulation) we do see that substantially more read-bandwidth is needed for the MCS solution relative to PVS. However, latency remained at the sub millisecond level throughout the entirety of all simulations. Full storage and host performance metrics of this test are available in the Design Guide.
From these results, we have found that Pure Storage will provide an outstanding end-user experience for either PVS or MCS technologies. With that being said, customers who are more focused on VM density per array will likely see a slightly better result using PVS over MCS due to the lower read-IOPs requirements associated with that technology.
As always, we encourage our customers to test with their own environments in order to best understand the implications of each solution to arrive at the optimal solution for their unique project.
We would like to invite you to our inaugural user conference in San Francisco March 14th and 15th where you will have the opportunity to engage and compare notes with your peers, talk to industry experts, see inspiring keynotes and game-changing product demos.
Find out more about Pure Accelerate here:
Lastly, this blog post is only a small excerpt from a much more comprehensive design guide where we show how easy it is to setup and scale your Citrix XenDesktop project while maintaining incredible performance from hundreds to thousands of desktops.
To download the Design Guide in its entirety please click on the below link.
Pure Storage, Pure1 and the “P” Logo mark are trademarks of Pure Storage, Inc. All other trademarks or names referenced in this document are the property of their respective owners.