Graphics testing in virtualized desktop environments: How-to size your launchers
One increasingly common way to support high-end applications is to implement GPUs on the host workstation in virtual desktop environments. Doing so accelerates hardware and makes it possible for geographically-distributed design teams to work collaboratively in real time via virtualized desktops. To meet this trend, Login VSI created the Login VSI Graphics Framework so that our customers can perform easy and repetitive testing specifically designed for graphics-intense applications. Since these GPU-enabled environments behave differently than “regular” virtual desktop environments, sizing of the Login VSI launcher machines is a bit different than regular Login VSI tests. In this blog, I will help you correctly size your Login VSI launcher machine for graphics testing.
For example, the Office Worker workload in Login VSI 4.1 has a limited amount of multimedia and heavy graphics applications. With the normal Login VSI workloads, like knowledgeworker and officeworker workloads, we recommend a maximum of 50 users per launcher when using RDP as connection protocol. For any other protocol, the recommendation is a maximum of 25 sessions per launcher. This is based on hardware of 2 CPUs and 4GB of RAM, which can be either virtual or physical hardware.
The Login VSI Graphics Framework is intended to be used with graphics-heavy applications. Citrix and NVIDIA introduced the vGPU on XenServer, which gives XenDesktop machines access to a virtualized GPU. Together with HDX 3D Pro, the vGPU technology proves to be very powerful for these types of rich graphics applications.
HDX 3D Pro compresses the graphical data it sends to the end client, which allows high-quality graphics with reduced bandwidth (bandwidth savings can be up to 50%). The compression takes CPU power on the guest operating system, which should be taken in account when sizing your GPU-enabled virtual desktop environment. The impact of HDX 3D Pro can be easily measured by using the Login VSI Graphics Framework.
The compressed graphical intensive content needs to be decompressed at the endpoint device. This decompression takes CPU power on the endpoint. When monitoring CPU usage of the Citrix Receiver on the endpoint, I measured an average CPU usage of 9%, with peak CPU usage going to 14%. This measurement was done on a laptop with a quad-core i7 processor.
The resolution of your endpoint device will make a big difference on the load that is generated on the virtual desktop environment. Higher resolution means more screen updates, which leads to higher CPU/GPU utilization.
Virtual hardware will not be able to use the resolution that a real-world end-user would use. You can think of multi-monitor configurations, which are not uncommon on CAD/CAM workstations. To be sure of the sizing of GPU-enabled virtual desktop environments, tests should be done which are as close to real-world as possible. This means that if your environment is configured with multi-monitor endpoints, you should test with multi-monitor launcher machines.
Only physical hardware supports multi-monitor configuration, and physical hardware is capable of higher resolutions. While it is absolutely possible to use virtual hardware as launchers in GPU-enabled virtual desktop environments, it is recommended to use physical hardware.
Taking into account the resolution and CPU utilization on the launcher machines, the maximum number of sessions launched from one launcher should be very limited. As stated before, the recommendation is to use physical hardware as launchers. The type of CPU installed will have a big impact on the maximum number of sessions. The recommendation is not to launch more than 5 sessions per launcher.
If possible, use a 1-to-1 mapping for launched sessions per launcher. This means that if you want to do a test with 30 users, using the Login VSI Graphics Framework you should use at least 6 launchers, but we recommend you use 30 physical launchers.