Author: Erik Bohnhorst, Solution Architect, Client Virtualization Engineering
For the midsized and enterprise markets, HP capitalizes on the increased performance and user density of the ProLiant DL380p Gen8 Server compared to the previous generation. HP has tested the ProLiant DL380p Gen8 with VMware View with Microsoft Remote Desktop Protocol (RDP) using LoginVSI - and the results are worthy of the ProLiant name: VSI Max score for medium script users – 226 users.
The following graph illustrates the CPU and memory utilization of 226 virtual desktops running the LoginVSI medium workload.
Measuring VDI Performance
HP used numerous tools to measure server performance for VDI (virtual desktop infrastructure). Those included using internally-developed, leveraging partners’ tools, and most recently, utilizing the newest version of LoginVSI which provides useful data when testing system performance.
LoginVSI is a benchmarking tool designed to measure the performance of centralized desktop environments such as Server Based Computing and Virtual Desktop Infrastructure.
All tests were performed on a single HP ProLiant DL380p Gen8 server; virtual machines (VMs) were optimized following VMware’s recommended procedures. The server under test featured 2 Intel XEON E5-2690 processors and 256GB of 1600MHz PC3-12800 RAM. The virtual machines ran Microsoft Windows 7 64-bit. Medium user VMs were assigned 1.5GB of memory. All VMs used a single vCPU. A dedicated linked clone model was used for testing.
LoginVSI offers a number of workloads to emulate different user behaviors. To best emulate HP’s traditional test methodology, the medium user workloads was selected. The workload represent productivity workers respectively. LoginVSI works by performing a series of workloads against a centralized or set of centralized workloads. It measures a series of predefined timings for actions that occur during these workloads (such as copies, application start times and compressing files). This data is evaluated in terms of number of VSI sessions and a VSImax score is output as a function of the baseline performance versus the response times of the actions under load.
HP starts launchers at a fixed, 25 second interval. All launchers are virtual machines connected to a domain and configured for NTP services. Tests are run multiple times to validate consistency as well as to check for problems in scripts. The best result of all tests is reported in this document. Prior to commencing a test run, the server under test (SUT) is rebooted and all virtual machines are restarted. The SUT is allowed a period of rest to reclaim memory prior to test start. This is generally a 2-3 hour period depending on the number of virtual machines being tested.
HP ProLiant DL380p Gen8 Results: Medium Workload – VSI Max of 226
The medium workload is frequently used to express user counts that represent general productivity users or “average” end users. HP has chosen to report this workload as being representative of such a user.
Using the ProLiant DL380p Gen8 server configured as listed in this document HP received a VSI Max score of 226 users with a medium workload. Response times for actions measured within the script remained very low as shown in the graph below validating the quality of HP’s end user experience. This is attributable to both the performance of the ProLiant DL380p Gen8as well as HP’s approach to storage for VDI on VirtualSystem CV2 outlined here.
Figure 2: Medium script VSI Max score.
A VSImax of 226 for the LoginVSI medium workload is an unmatched result in the industry which is based on the optimal configuration of the HP ProLiant DL380p Gen8 server with its 2 Intel XEON E5-2690 processors and 256GB of memory running at 1600MHz. The graphs below show that CPU utilization is at 100% when VSImax is reached where as memory utilization is at 60% (155GB).
Real world estimations and recommended sizings
It has been HP’s experience over many years of working with customers on VDI implementations that the quality and consistency of the end user experience is a large factor in determining end user acceptance as well as long term success. Table 1 shows the VSIMax as well as the recommended planning numbers which are factored as 60-65% of the VSImax score.
Table 1. Recommended planning numbers by user type.
Figure 3: CPU utilization
Figure 4: Memory utilization
Real world sizing possibilities
A VSImax of 226 for the LoginVSI medium workload was achieved with 1.5 GB of memory per user. The LoginVSI medium user workload that includes Outlook 2010, Excel 2010, Word 2010, Adobe Acrobat Reader, Adobe Shockwave, Internet Explorer, Flash Videos and other applications has shown that 1.5 GB of memory is sufficient memory for medium users. This is the minimum HP recommends for a VM running Windows 7.
At the HP recommended user counts the VMs can be run with 2GB or more without affecting system performance. Below graphs show CPU and memory utilization for 147 LoginVSI medium workload users with 2 GB of memory.
Figure 5: CPU utilization at 147 users (2GB VMs)
Figure 6: Memory utilization at 147 users (2GB VMs)
The reference architecture sizing is derived from the medium script load with the capabilities of the storage to handle 18 IOPs per user in a 50/50 RW ratio. This storage requirement is derived from fibre channel traces captured by HP of running VDI environments. Native storage read write ratios in production have a large range of variability. It is important that you understand and assess your individual environment to properly size. Management VMs with the exception of those noted remain within the same set of servers as end user computing resources and will have an effect on overall server sizing. Resources are left available to handle system level failures as well as maintenance tasks.
There are many variables that will have an effect on your overall sizing including proper image optimization, memory speed and architecture, concurrency ratio of active users to total users, application delivery methods, operating system choice, protocol and provisioning methodology. All sizing numbers should be used as guidelines only.
For more information on HP client virtualization solutions, visit www.hp.com/go/cv
For more information on HP Gen8 servers, here