The Effect of Memory Speed on VDI User Density
Wed, 19 Aug 2020 23:34:48 -0000
|Read Time: 0 minutes
In most modern-day virtual desktop infrastructure (VDI) deployments, RAM (often referred to as memory) is not a bottleneck. More often, it is the processor that gets saturated before memory and storage does. However, it is not recommended to overcommit memory capacity for VDI deployments. It is important that there is a balance between the memory required by virtual machines and host physical memory. Low memory allocation can cause increased storage I/O due to excessive paging. Conversely, if RAM allocation is too high, it affects storage capacity negatively due to the increased size of page files, virtual machine swap files and suspend files.
Memory speed or speed of the memory bus is one of the other attributes of RAM, apart from memory capacity, that may affect the performance of your VDI system. The Dell EMC Ready Solutions for the VDI team recently completed some performance analysis work to check the impact of memory speeds on the ‘density optimized’ configuration offered as a part of Solutions for VDI. The density optimized configuration is based on Intel Xeon Scalable 2nd generation processors code-named Cascade Lake. In this blog, we will discuss the details of this performance analysis work to understand the effects of memory speed on VDI system performance.
Test environment
The VDI Engineering team performed tests with Login VSI, an industry standard tool for benchmarking VDI workloads. The tests were done using Login VSI Knowledge Workload running on VMs configured with 2 vCPUs and 4 GB of RAM with that 4 GB of RAM all being reserved memory.
The testbed environment was a 3-node cluster of VxRail V570F appliances that was optimized for VDI workloads. The cluster was configured and tested with 768 GB of memory per node with a speed of 2666 MHz for test1 and a speed of 2933 MHz for test2. The Environment configuration used was:
- PowerEdge R740xd servers
- Intel Xeon Gold 6248, 2 x 20-core, 2.5 GHz processors
- 768 GB memory (24 x 32 GB @ 2666 MHz) (2 DIMMS per channel (DPC) for test1)
- 768 GB memory (12 x 64 GB @ 2933 MHz) (1 DIMM per channel (DPC) for test2)
- vSAN hybrid data store using an SSD caching tier
- VMware ESXi 6.7 hypervisor
- VMware Horizon 7.7 VDI software layer
The compute workload virtual machines were Windows 10, 64-bit, version 1803. One of the VxRail cluster nodes hosted both management and compute virtual machines. The other two nodes were dedicated to workload compute. Figure 1 shows the main components involved in this work.
Figure 1 Dell EMC VxRail Solutions for VDI Stack Components
Results Analysis
Now let’s check the Login VSI results from the tests done with memory speeds of 2666 MHz and 2933 MHz. Figure 2 shows the comparison graphs of the Login VSI Index Average values (the average response time for the system). From the graphs, we can see that the difference in response times from the two tests was marginal while sessions were loaded. We can ignore these marginal differences when doing a Login VSI test that is based on random workloads. While reaching a CPU utilization threshold of approximately 85%, active session count was 480 from both tests, implying that memory speed doesn’t affect user densities significantly in a VxRail density optimized configuration based on Intel Xeon Scalable Gold 6248 processors.
Note that the Dell EMC Ready Solutions for VDI team considers 85% of CPU utilization as a threshold because testing and loading the system beyond this value might have a negative impact on the performance and end-user experience. So, in these tests, the system was not stressed to the point of reaching a Login VSIMax. VSIMax shows the number of sessions that can be active on a system before the system is saturated.
Figure 2 Login VSI response time comparison with different memory speeds 2666 MHz vs. 2933 MHz
Login VSI test results metrics are summarized in Table 1 below.
Table 1 Login VSI Test Summary
Figure 3 shows the comparison of processor utilization in tests done with memory speeds of 2666 MHz and 2933 MHz. As shown in the figure, we couldn’t see a notable difference in the processor utilization in these tests. CPU utilization steadily increased during the login phase in both tests. The test with 2933 MHz showed a comparatively lower utilization, however, the difference was marginal. The difference in steady-state average CPU utilization was around 4% in these tests.
Figure 3 Comparison of CPU utilization with 2666 MHz and 2933 MHz memory speed
Summary
To summarize, our tests showed that in a VDI system based on the Dell EMC VxRail Density Optimized configuration powered by Intel Xeon Scalable Gold 6248 processors, an increase in memory speed did not improve the overall performance of the selected application workload significantly. It was also evident from our testing that memory was never a bottleneck during the testing. We did not test with other processor models. The results might vary when tested with other models.
In the next blog, we’ll discuss the effect of different Microsoft Windows operating systems versions on VDI user density. So, stay tuned!