You must have seen the announcements around Citrix and Cisco partnering together to deliver a joint comprehensive solution for desktop virtualization. This is very exciting and there is a lot of industry buzz around it. Being the Citrix Consultant onsite at Cisco along with Frank Anderson from the XenDesktop Product Marketing team, I would like to provide technical insight in to the published Validated Design Guides to help provide understanding around the testing approach taken and the scalability results obtained.
The Reference Architecture focuses on providing a validation and scalability study of a Citrix XenDesktop environment on Cisco UCS B-Series blade Servers connected to the NetApp Storage Array. For this, the FlexCast TM models of Hosted VDI and Hosted Shared were designed and implemented at Cisco.
For the Hosted VDI model, the solution scales from a single-Chassis containing a single Cisco UCS B250 M2 blade up to a Four-Chassis containing a total of 16 blades in a linear fashion. The Scalability numbers are very impressive which provides a really good value proposition for customers. The validated density was achieved based on the following analysis: Each Windows 7 virtual deskttop was configured with 1.5GB of RAM and each Cisco UCS B250 M2 blade was configured with 192GB of RAM (UCS B250 M2 Blades can support up to 384GB RAM). This allowed the placement of 110 virtual desktops per blade with ~85% of the available memory utilized while accounting for the hypervisor overhead (for both Citrix XenServer and VMware vSphere). This provided the following linearly scalable solution for the Hosted VDI model:
|Number of Cisco UCS Blades||Number of Virtual Desktops|
|Single blade (Single-Chassis)||110|
|8 blades (Two-Chassis)||880|
|16 blades (Four-Chassis)||1760|
For the Hosted Shared model (XenApp sessions), single server scalability tests were conducted on Cisco UCS B200 M2 Blade with 92GB of memory. The goal of this testing was to find an optimal virtualization configuration for multiple XenApp virtual machines on a single blade and for this multiple VM configurations were tested on a single blade. The optimal results are as shown in the graphs below:
Figure 1 Optimal XenApp VM configuration on Citrix XenServer
Figure 2 Optimal XenApp VM configuration on VMware vSphere
For this Xenapp Single Server Scalability, the optimal scalability numbers were determined by the percentage of CPU utilization on the hypervisor while keeping the workload response times acceptable as per the Login VSI benchmarking methodology. This is different than the memory based calculation done while placing virtual desktops for the Hosted VDI model. Hence in this case, the number of XenApp sessions supported per blade has a potential to differ between hypervisors and it does. The maximum number of XenApp sessions supported on a single Cisco UCS B200 M2 blade being 180 for Citrix XenServer and 120 for VMware vSphere.
The user workloads executed for all project testing were Login VSI based simulated workflows closely representing real-world user activities and the response times were measured with the goal of finding the number of sessions the system can handle without degrading user performance.
Let me be clear that the goal of this validation was by no means to conduct a hypervisor comparison, but it is pretty obvious to note that Citrix XenServer does a pretty good job (better) of executing XenApp workloads on Cisco UCS Blades.
For a complete overview of the project and testing, please refer to the Cisco Validated Design Guides here
Principal Consultant, WW Consulting Solutions