Solutions based on Dell EMC VxRail provide flexibility as you scale, reducing the initial and future TCO. Add additional physical and virtual servers to the server pools to scale horizontally. Add additional resources such as faster CPUs, more memory or more disk capacity to scale vertically. The following figure shows an abstract view of how the solution can scale as you add VxRail to the solution to support additional users and AI virtual machines when supporting a mixed workload.
In the parallel use case, when the environment is reaching peak VDI user utilization, the workload orchestrator can power off the AI VMs and bring up additional VDI VMs to support the user workload at that time. This is demonstrated in the following figure where all the compute and related GPU resources are dedicated to the Virtual Workstation VDI workload.
Scaling out
Each component of the solution architecture scales independently, depending on the required number of supported users. You can add nodes at any time to expand the vSAN SDS pool in a modular fashion. The scaling limit for a vSphere cluster is restricted by the limits of vSAN at 64 nodes per block.
The boundary for a Horizon block is the vCenter. The number of virtual machines a vCenter (and therefore a block) can host depends on the type of Horizon 8 VMs being used. The recommended limits for a Horizon block at the time of writing are as follows:
- 12,000 full clone VMs
- 12,000 instant clone VMs
- 4,000 linked clone VMs
For the latest sizing guidance, see VMware Configuration Maximums. Additionally, see the VMware Knowledge Base article VMware Horizon 7 sizing limits and recommendations (2150348).
This design guide presents the use of instant clones in the following figures. We used design limits of 4,000 instant-clone VMs per block and up to 12,000 VMs per pod. VMware Horizon pools have a limit of 4,000 VMs, so additional pools are necessary when scaling above that.
The VMware Horizon management infrastructure and Knowledge-User VMs are located on separate vSphere clusters. Four management nodes are a suitable configuration to start with to provide redundancy and self-healing and can be scaled as appropriate. This logical model can be implemented using VMware Validated Designs or using VMware Cloud Foundation.
The following figure shows a 4,000-user pod supporting up to 4,000 Knowledge-User VMs with a single resource block and two vSphere clusters. With the above limits in mind, 56 compute nodes with 72 Knowledge-User VMs per node across two vSphere clusters would reach the maximum number of VMs for the block.
The following figure shows a scale-out to a 12,000-user Horizon pod with three 4,000-user resource blocks and six vSphere clusters.
Scaling up
Dell Technologies recommends a validated disk configuration for general-purpose VDI. These configurations leave drive slots available for future vertical expansion and ensure that you protect your investment as new technology transforms your organization.
For more information about Horizon pod/block architecture and scaling, see the VMware Workspace ONE and VMware Horizon 7 Enterprise Edition On-premises Reference Architecture Guide.