Running a Dell Cloud Client Computing Environment in a Boot-On-Demand Environment

By John Kelly and Yinglong Jiang

Powering and cooling represent a large component of datacenter costs. In an ideal scenario, it is possible to minimize these costs by optimizing the datacenter’s energy consumption without impacting performance. The desire to meet these dual requirements has implications across the entire spectrum of enterprise IT and desktop virtualization is no different.

In this article we present the results of a performance analysis activity carried out on a Dell DVS Enterprise with VMware Horizon View environment to assess the impact of various boot-on-demand scenarios on resource utilization and end-user experience and the implications of these scenarios for performance and power optimization

One particular area of relevance from a desktop virtualization perspective is desktop pool management: virtual desktops must be available for users when they need them, while optimizing power utilization when they don’t use them – and any boot activities that occur in the transition phase while desktops become available must not affect the holy grail of desktop virtualization, end-user experience.

Some commonly used software-based approaches to the above include:

1. VMware View “Number of spare (powered on) desktops” setting.
2. XenDesktop “Idle Desktop Count” management based on time of day.
3. VMware “Host Power Management” and “Distributed Power Management” settings.
4. Powershell (e.g. PowerCLI) based desktop power management.

The above approaches can be used either individually or together to provide the optimal power consumption / virtual desktop availability combination. However, many of these approaches involve a so-called “boot-on-demand” environment, where less than 100% of desktops in a pool are powered on at any time, with subsequent desktops typically being powered on to keep a minimum number of desktops available for logins. But this has obvious implications for performance, since the related boot activities will cause significant server and storage activity over and above what would be seen if all desktops are available at all times.

In order to carry out the performance analysis activity described above, 2 representative boot-on-demand environments were used. The environment was a standard configuration Dell desktop virtualization with VMware Horizon View environment i.e. Dell PowerEdge R720 server, 192GB RAM, 2.9GHz processor, VMware Horizon View 5.2. The pre-boot configurations tested were:

1. 116 user desktop pool, Login VSI Medium (Dell standard) workload, 20% of desktops pre-booted.
2. 116 user desktop pool, Login VSI Medium (Dell standard) workload, 5% of desktops pre-booted.

The pre-boot configurations described above were selected as being representative of appropriate levels of pre-booting/boot activity, with the 5% pre-booted environment generating a particularly aggressive level of desktop booting. It should be noted that, based on Dell  performance analysis and characterization activities, 116 standard users is the current per-server density for a 100% pre-booted environment with the above configuration and so this serves as a useful comparison point for the 5% and 20% pre-booted configurations.

Table 1 shows a comparison of a number of storage IO related parameters for the 5%, 20% and 100% configurations.

Table 1: Storage IO Parameters of Interest for Pre-Boot Scenarios

It is important to note a number of key elements of Table 1 above:

• As expected, IOPS per user increases as the percentage of pre-booted desktops decreases, due to the increased disk IO activity caused by increased levels of desktop boot activity.
• The read-write percentage varies from a ratio of 22/78 (read / write %) to 35 / 65 at 20% pre-booted and 37 / 63 at 5% pre-booted.  The 22/78 number is similar to the commonly quoted 20/80 ratio for VDI steady-state usage, while the increasing percentages of read activity as the percentage of pre-booted desktops reduces is as expected, since desktop booting is a read-intensive operation from a disk IO perspective.
• Maximum latency across the pre-boot configurations never exceeds 7milliseconds. This is probably the single most important result of this testing, since this is well below the broadly accepted 20ms performance threshold that is commonly used in the storage industry as an indicator of a disk IO bottleneck; a maximum latency of 7ms is therefore regarded as delivering good end-user experience from a storage IO perspective.

Figure 1 below shows a comparison of CPU utilization during a Login VSI test for the 3 pre-boot scenarios. CPU utilization is typically the limiting factor for per-host desktop density in the environment under discussion, so a similar level of CPU utilization will result in an unchanged per-host density number. It can be seen that CPU utilization is indeed broadly similar for each scenario, indicating no reduction in per-host density, independent of pre-boot case chosen.

Figure 1: CPU Utilization Comparison for 5%, 20% and 100% Pre-Boot Configurations

In addition to the CPU and storage utilization information presented above, the Stratusphere UX end-user experience monitoring tool was also used during this performance analysis activity for each of the pre-boot configurations under discussion. Results from this tool indicated that the good end-user experience observed during 100% pre-boot testing was preserved with the 5% and 20% pre-boot configurations. This was as expected, based on storage and CPU utilization results shown in Table 1 and Figure 1 respectively.

The performance analysis activity described above illustrates that Dell and VMware solutions for desktop virtualization are robustly designed for multiple pre-boot configurations and can deliver a good end-user experience across these multiple pre-boot configurations. Customers should consider a boot-on-demand environment where datacenter power utilization is a concern.

Dell and VMware solutions for desktop virtualization deliver secure and predictable computing. It is, of course, advisable to engage with Dell for a comprehensive analysis of your environment, including activities such as a Blueprint Assessment and Proof-Of-Concept, in order to ensure an optimized deployment of a Dell desktop virtualization solution in your organization.

Additional resources:

Dell DVS Enterprise with VMware Horizon View 5.2 Reference Architecture

Dell Desktop Virtualization Solution web site