Virtual desktop infrastructure is unpredictable for one simple reason: People are unpredictable.
A server in the data center is built to suit a consistent task that really only changes when something modifies the application using it. Human beings use VDI, however, and it's impossible to know how they will behave. Add in the complex technical underpinnings, and it's obvious VDI issues force IT to troubleshoot in new ways.
VDI problems generally fall under one of four categories: persona management, performance, infrastructure and platform. These issues often overlap. An infrastructure problem might easily affect performance, for example, but each issue does have its own characteristics.
VDI troubleshooting often starts with users complaining their VMs load slowly, or they're missing files. Whether IT uses native persona tools in VMware Horizon or runs Liquidware Labs Profile Unity, Unidesk, AppSense, or Active Directory Group Policy, there are a few questions they should ask.
In most cases, personas involve data redirection, temporary files and settings stored to a network volume or an attached data store at the VM level. So, if users are complaining, it's a great idea to draw out the connectivity architecture on a whiteboard.
If the redirection sits on network attached storage (NAS), is the network between the NAS and the VM adequate to rapidly move user data during login? What about the underlying storage for attached data stores? Do those volumes have enough overhead for the spike in throughput and IOPS when users log in or out of a VM?
If users complain about missing applications, VDI shops must ask if they really have the right product for departmentally installed and user installed applications. Does their persona management product even support that requirement? If not, it might be time to look at other options.
Another indicator that persona management is inadequate is if users experience problems accessing their saved customizations or apps after IT recomposes a virtual desktop pool.
Virtual desktop performance issues, such as supporting computer-aided design, 3D modeling and graphics and video rendering, are another source of user strife. These complaints generally come from a localized group of users such as designers, engineers, marketing professionals and some executives with greater system requirements than most of the user population.
It's not a bad idea to isolate specialized user groups in dedicated high-performance pools with a master image tuned for their specific use cases. Offer increased amounts of virtual memory on these VMs and increase the number of monitors and resolution to the maximum setting on the pool. This provisions the VMs with the maximum video memory allocation.
IT must not overassign virtual processors if they create a high-performance master image. As tempting as this is, more often than not it decreases performance due to CPU core scheduling wait times. If admins already assigned more than two virtual CPUs to desktop VMs, it might be time to rethink their strategy. Also, rethink using any CPU Affinity restrictions, because it does not dedicate a core to a specific VM, which hurts performance.
If a VDI deployment is large enough, IT can cluster high-performance users together. Enabling clusters with hardware-based protocol accelerators (APEX offload cards) or hardware GPUs (NVIDIA Grid) can significantly improve virtual desktop performance.
At an infrastructure level, VDI troubleshooting often revolves around resource constraints. Examine the deployment at high-load times, looking for CPU, memory, storage and switching saturation on the hosts.
Speed is key for VDI users.
A 10 GbE is really a mandatory requirement between VDI hosts and any storage array serving them. If the array doesn't support high levels of throughput and IOPS loads, or 10 GbE is not an option, IT may consider other options to increase infrastructure performance. For example, deploying solid-state drives in the host and enabling hypervisor-level caching could result in a significant change in storage and network performance.
An often overlooked issue on the infrastructure side is users' endpoints. For example, some older thin clients don't have the horsepower to run the newer broker software and will show Out of Memory errors. In addition, USB redirection technology often doesn't work with older devices.
Platform and auditability concerns
On the platform side of VDI troubleshooting, it's important to maintain version consistency between the VM's built-in tools, the brokers and the clients. Issues often end up boiling down to a problem where IT upgraded the VMware Horizon version, for example, but nobody updated the new VMware Tools version or Horizon View agent. Likewise, it is routine for everything on the server side to look great, but some endpoints run a client that's three revisions behind.
One final thing to consider is auditability. Most IT departments have performance measurement tools that monitor their data center infrastructure. For some reason they don't always apply the same tools to VDI deployments even though the tool sets are available. Companies such as Liquidware Labs with its Stratosphere UX product, Manage Engine and SolarWinds offer third-party tools for VDI performance management. Additionally, most VDI software providers, including VMware, Citrix and Microsoft offer a monitoring tool for their products.
How to fight common VDI problems
VDI questions IT should expect from users
What will users want to know about VDI?