This section provides best practices for sizing your VDI deployment.
Platform configurations
With several configurations to choose from, consider these basic differences:
- The Density Optimized configurations provide a good balance of performance and scalability for various general-purpose VDI workloads.
- The Virtual Workstation configurations provide the highest levels of performance for more specialized VDI workloads, which means you can use it with ISV and high-end computing workloads.
CPU
User density and graphics considerations include:
- For architectures with 2nd Generation Intel Xeon Scalable processors:
- Knowledge Workers—2.8 users per core. For example, 44 knowledge users with dual eight-core processors
- For architectures with 3rd Generation Intel Xeon Scalable processors:
- See the VDI Design Guide—VMware Horizon on VxRail and vSAN Ready Nodes for approximate scaling guidance.
- For graphics:
- For high-end graphics configurations with NVIDIA vWS graphics enabled, choose higher clock speeds over higher core counts. Many applications that benefit from high-end graphics are engineered with single-threaded CPU components. Higher clock speeds benefit users more in these workloads.
- For NVIDIA vPC configurations, use higher core counts over faster clock speeds to reduce oversubscription.
- Most graphics configurations do not experience high CPU oversubscription because vGPU resources are likely to be the resource constraint in the appliance.
Memory
Best practices for memory allocation and configuration include:
- Do not overcommit memory when sizing, because memory is often not the constraining resource. Overcommitting memory increases the possibility of performance degradation if contention for memory resources occurs (for example, swapping and ballooning of memory). Overcommitted memory can also impact storage performance when swap-files are created.
- 3rd Generation Intel Xeon Scalable Processors—Populate memory in units of eight DIMMs per CPU to yield the highest performance. Dell PowerEdge servers using 3rd Generation Intel Xeon Scalable processors have eight memory channels per CPU, which are controlled by four internal memory controllers, each handling two memory channels. To ensure that your environment has the optimal memory configuration, use a balanced configuration, where each CPU supports a maximum of 16 DIMMs (or 32 DIMMs for a dual-CPU server). The most effective configuration on the PowerEdge R650 and R750 is 16 DIMMs (8 per processor).
- 2nd Generation Intel Xeon Scalable Processors—Populate memory in units of six per CPU to yield the highest performance. Dell PowerEdge servers using 2nd Generation Intel Xeon Scalable processors have six memory channels per CPU, which are controlled by two internal memory controllers, each handling three memory channels. To ensure that your environment has the optimal memory configuration, use a balanced configuration, where each CPU supports a maximum of 12 DIMMs (or 48 DIMMs for a quad-CPU server). The most effective configuration on the PowerEdge R840 is 24 DIMMS (6 per processor).
- Use Intel Optane Persistent Memory (PMem) for cost savings over traditional DRAM or in situations where high memory capacity is required (1 TB or greater). vSphere 7 update 3 introduced vSphere Memory Monitoring and Remediation (vMMR), which provides visibility of performance statistics for tiered memory. For additional information, see the VMware documentation on vMMR.
NVIDIA vGPU considerations
- vPC licenses that support up to 2 GB of frame buffer and up to two 4K monitors or a single 5K monitor to cover most traditional VDI users. Maximum node density for graphics-accelerated use can typically be calculated as the available frame buffer per node divided by the frame buffer size.
- The addition of GPU cards does not necessarily reduce CPU utilization. Instead, it enhances the user experience and offloads specific operations best performed by the GPU.
- Dell Technologies recommends using the BLAST protocol for vGPU enabled desktops. NVIDIA GPUs are equipped with encoders that support BLAST.
- Virtual Workstations are typically configured with at least 2 GB video buffer.
Sizing considerations
Best practices for sizing your deployment include:
- User density—If concurrency is a concern, calculate how many users will use the environment at peak utilization. For example, if only 80 percent are using the environment at a time, the environment must support only that number of users (plus a failure capacity).
Disaster recovery—For DR planning, Dell Technologies recommends implementing a dual/multi-site solution. The goal is to keep the environment online and, in case of an outage, to perform an environment recovery with minimum disruption to the business.
- Management and compute clusters—For small test environments, it is acceptable to use a combined management and compute cluster. For environments deployed at a larger scale, we recommend that you separate the management and compute layers. When creating a management cluster for a large-scale deployment, consider using the PowerEdge R650 platform to reduce the data center footprint. With a larger variety of configuration possibilities, the PowerEdge R840 or R750 platforms are preferred for compute clusters.
- Network isolation—When designing for larger-scale deployments, consider physically separating the management and VDI traffic from the vSAN traffic for traffic isolation and to improve network performance and scalability. This design illustrates a four-NIC configuration per server with all the traffic separated physically as well as logically using VLANs.