Citrix just released the XenDesktop Planning Guide – XenDesktop Scalability and these are the results (also from the PDF):
| Desktop Controller | SQL Database | |||||
| Type | Physical server | Physical server | ||||
| Operating System | Windows 2008 R2 | Windows 2008 R2 | ||||
| Product Version | XenDesktop 5 | SQL Server 2008 SP1 | ||||
| # of CPU | Dual, quad core 1.86Ghz | Dual, quad core 1.86Ghz | ||||
| Memory | 16 GB RAM | 16 GB RAM | ||||
| Disk | Single SAS drive (10,000 RPM) | Fibre Channel | ||||
| Vol 1: SQL temp DB | ||||||
| Vol 2: SQL install and DB | ||||||
| Number of servers | 2 servers | 3 servers (Principal, Mirror, Witness) | ||||
| Configuration Information | Power management disabled | Synchronous database mirroring with automatic failover | ||||
| Desktop Load | 20,000 simulated desktops | 20,000 simulated desktops | ||||
| Boot Storm | 40-50% CPU utilization during the virtual desktop registration process | 15-25% CPU utilization across the SQL principal database | ||||
| (20,000 desktops in 10 minutes) | 5-10% for SQL mirror | |||||
| The SQL witness was essentially idle. | ||||||
| XenDesktop Controller | SQL Database | ||
| Type | Virtual | Physical server | |
| Operating System | Windows 2008 R2 | Windows 2008 R2 | |
| Product Version | XenDesktop 5.0 | SQL Server 2008 SP1 | |
| # of CPU | 2 vCPU | Dual, quad core 1.86Ghz | |
| Hypervisor host: Dual, Quad core 2.26Ghz | |||
| Memory | 4GB RAM | 16GB RAM | |
| Disk | 30GB image on NFS share | 34GB Local Hard disk | |
| 400GB FC disk (Database) | |||
| Number of servers | 2 servers | 3 servers (Principal, Mirror, Witness) | |
| Configuration Information | Power management disabled | Synchronous database mirroring with automatic failover | |
| Desktop Load | 2,500 desktops | 2,500 desktops | |
| Boot Storm | 10% CPU utilization during the virtual desktop registration process | No data available | |
| (2,500 desktops in 4 hours) | |||
| Logon Storm | 12-15% CPU utilization during user connection process | No data available | |
| 2,500 connections in 30 minutes) | |||
| User Perception | Launch and logon operations averaged 2.5 seconds for all 2,500 users | ||
The key when planning the XenDesktop site configuration is to follow these recommendations:
1. N+1: Always make sure there is extra capacity in the event that a single controller fails. If two controllers are required to support the user/desktop load, three should be implemented.
2. Estimate: Based on the boot storm, logon storm and hardware selected, create a rough estimate based on the above examples. The following can be used as a rough estimate when CPU resources are dedicated to the controller:
a. Desktop registrations per minute per dedicated core: 125-180 desktops
Assumes the desktops are delivered via Provisioning Services and that the controller’s CPUs are not shared with other virtual servers.
b. Logons per minute per dedicated core: 100-120 users
Over-committing a controller or SQL database will make the initial user launches slower than if the resources are under-committed. Additional resources can be added easily to overcome estimation discrepancies.
Sharing vCPUs will have an impact on overall scalability as those cores are not 100% dedicated to controller activities. Care must be taken when virtualizing controllers with other virtual servers.
3. Monitor: The most important thing is to do is to monitor the resources. The following are recommended metrics to observer the Controller scalability as well as the user experience:
XenDesktop Controller CPU Process Utilization Citrix XML Service Enumerate Resources (Avg Transaction Time)Validate User Credentials (Avg Transaction Time) Launch Get Address (Avg Transaction Time)
Launch Get Logon Ticket (Avg Transaction Time)
Note: Disk and network utilization were negligible during the tests for the XenDesktop Controllers.
As we can see the Desktop Controller is very scalable and provides good leverage to design and build large farm without performance degradation. Keep in mind that these are tests in a controlled environment and should just be used as guide lines, so testing should be done at site and depending on the customer configuration and requirements.
Kees Baggerman
Latest posts by Kees Baggerman (see all)
- Nutanix AHV and Citrix MCS: Adding a persistent disk via Powershell – v2 - November 19, 2019
- Recovering a Protection Domain snapshot to a VM - September 13, 2019
- Checking power settings on VMs using powershell - September 11, 2019
- Updated: VM Reporting Script for Nutanix with Powershell - July 3, 2019
- Updated (again!): VM Reporting Script for Nutanix AHV/vSphere with Powershell - June 17, 2019
