End-To-End Network Bandwidth Utilization in a VDI Environment

By John Kelly and Andrew Breedy.

The consolidated nature of a virtual desktop infrastructure (VDI) environment has led to significant discussion around where the hardware resource bottlenecks lie in such an environment, since the aggregation of resource utilization across multiple users and virtual desktops has the potential to create differing resource utilization profiles. Storage IO requirements, for example, have consistently been a widely discussed topic in the VDI domain, while Dell Desktop Virtualization Solutions (DVS) has consistently seen CPU as the resource bottleneck which is reached first. However, network bandwidth resource utilization is also an important topic when considering both the initial migration to a VDI environment and the ongoing health of the environment. Read on for information on why this is the case, how usage can be monitored using Dell OpenManage Network Manager (OMNM) software provided by Dell and some representative information based on performance analysis work carried out in Dell labs.

The main additional network traffic present in a VDI environment is display protocol traffic i.e. the network traffic that carries the images, text etc., seen on the user’s display device from the virtual desktop in the corporate datacenter to the location of the display device. Typically, conversations in relation to VDI network utilization have focused on this display protocol – questions such as which is the “best” display protocol, what is typical display protocol bandwidth utilization etc. are common discussions in the VDI domain. From the perspective of display protocol bandwidth utilization, figures from 30 kb/s for WAN-optimized task worker usage up to 2 Mb/s for non-optimized HD video are commonly discussed. However, it is important to also be conscious of other network traffic in a VDI environment, as shown in Figure 1 below.

Figure 1: Components of Network Utilization in a VDI Environment

The network traffic shown in Figure 1 (resulting from normal user activities in a corporate environment) will generate significant traffic in the datacenter that has a different pattern than is the case when physical desktops are located at users’ desks. The consolidated nature of a VDI environment discussed above results in the consolidation of this traffic and it is important to ensure that significant network capacity is available to handle this traffic. The Dell DVS engineering organization has carried out some performance analysis work in a representative VDI environment to help with this capacity planning – but first a brief introduction to the network monitoring tools used: Dell OpenManage Network Manager.

OMNM’s centralized management solution for Dell networking environments provides discovery, configuration management, monitoring and reporting for the entire Dell Networking family of products (PowerConnect and Force10). OMNM automates the discovery of network devices, and provides detailed information on the devices and their connectivity, including the ability to draw physical and logical topology maps. It also enables the network administrator to monitor the health and performance of their Dell network, allowing the creation of dashboards to capture important events and trends, and display them over time. From the perspective of the current activity, OMNM uses industry standard device management and monitoring technologies such as SNMP and sFlow to facilitate the discovery and monitoring of network devices.

The key components of the system-under-test (SUT) VDI environment used for this performance analysis activity are shown below:

• Microsoft VDI environment, based on Windows Server 2012 R2, with the relevant hypervisor (Hyper-V) and desktop virtualization (RDVH) roles enabled.
• Dell R720 server hardware used for both the compute host (i.e. where the virtual desktops run) and the management host (i.e. where the various RDVH management roles run) – so 2 * R720s in total. Local host storage was used throughout.
• Login VSI 4 VDI load-testing software, using medium user workload with a density of 75 virtual desktops per compute host; the medium workload is representative of the knowledge worker use case.
• Dell Powerconnect 6248 top-of-rack (TOR) switch and Dell Force10 S4810 core switch.
• Both management and compute hosts use teamed 1Gb NICs for each VLAN (i.e. management VLAN and VDI VLAN) with the uplink connectivity from the TOR switch to the core switch being delivered by the aggregation of 2 * 10Gbps links.

It is important to note in the context of the above components that the purpose of this blog is solely to look at network utilization in a representative VDI environment and not to present a reference architecture.

The network configuration of the SUT used for this work is shown in more detail in Figure 2 below. The VLAN usage shown is a standard VDI network configuration

Figure 2: SUT Network Configuration

So now that we’ve described the information that we need and the configuration that’s being used to gather that information, it’s time to present the information gathered using OMNM. Since network capacity planning guidance is the primary function of this blog, the information presented in Table 1 below is presented on a per-user basis i.e. bandwidth requirement per user, based on maximum bandwidth utilization during Login VSI load-testing.

Table 1: Per-User Bandwidth Requirement

It can be seen from Table 1 above that a typical Login VSI 4 medium user generates just under 2,500kb/s of traffic on the VDI VLAN shown in Figure 2 above, with a slightly larger amount of traffic (just under 3,000 kb/s) being generated on the core uplink; this additional traffic is composed of traffic between the management host (where the RDVH user profile disks and RDVH management roles reside) and the core. Traffic on the management VLAN is quite small. The right-most column in Table 1 gives indicative information on the number of users that can be hosted by a 1Gbps NIC for each of the network connection points. This should not be regarded as a network design recommendation (e.g. teamed NICs will usually be used for load-balancing and failover purposes and the core uplink configuration of 2 * 10Gb aggregated links shown in Figure 2 is a more widely-used core uplink configuration) but nevertheless it provides useful guidance.

There are a number of important takeaways from the above information:

1. Don’t be tempted to regard figures you see for display protocol usage as representative of overall virtual desktop network utilization – non display protocol traffic still exists (like it does for any system) and will contribute significantly to network load.
2. Two aggregated 10Gb uplinks from a top-of-rack switch to the network core should have the capability to host significant numbers of users – between 5,000 and 10,000 based on the above numbers. Of course, if you have any other non-VDI network traffic on the core uplink, graphics-intensive VDI users etc., this will impact capacity.
3. As is the case for all VDI implementations, your mileage may vary! Engage with the Dell Global Infrastructure Consulting Services (GICS) team to blueprint your physical environment, then run a Proof-Of-Concept in a Dell Solutions Centre in order to get full visibility into your environment.

The information presented above provides a useful starting point when considering network traffic in a VDI environment: by taking into account the information shown and carefully planning your implementation based on the guidance above, your chances of a successful VDI implementation will be significantly increased.