agsandrew - Fotolia
Top 4 challenges of hyper-converged infrastructure for VDI
HCI and VDI often go well together, providing simpler scalability and flexibility for virtual desktops. But what happens when a VDI deployment has requirements outside the norm?
Hyper-converged infrastructure is an excellent fit for desktop virtualization, but not all VDI deployments are the same. There are a few limitations to HCI that might make it a poor fit for an organization's VDI if the requirements are not average.
Some of the common, but potentially limiting, HCI features include the use of high-density, dual-socket servers and providing storage availability through a multinode cluster. Just as VDI deployments have different requirements, HCI offerings from different vendors have differing capabilities. An unusual VDI deployment may require an unusual HCI platform.
Get to know four of the top challenges IT may encounter when using hyper-converged infrastructure with VDI in special cases.
Dual-socket servers bring some limitations
All of the physical appliance HCI vendors use dual-socket servers for their HCI nodes, and most use four to 12 CPU cores for their storage software. The result is around 20 CPU cores per node for running desktop VMs, which is plenty if IT can run seven to 12 virtual CPUs (vCPUs) per core.
Limitations arise with hyper-converged infrastructure for VDI in organizations that have CPU-heavy applications, particularly where each desktop VM needs two or more vCPUs. On an HCI platform, IT might only be able to run 20 of these VDI desktops per host, where a quad-socket host connected to a storage area network (SAN) might triple the number of desktops per physical server.
Remember that desktops per host is not the critical measure; IT wants to know the cost per desktop. An HCI deployment with two or three times the number of nodes might still be more cost-effective than fewer but larger physical servers and a SAN.
Compact servers limit GPU power
Many HCI deployments use server sleds in 2U enclosures. These sleds usually have only one or two Peripheral Component Interconnect Express slots for expansion, and most do not have sufficient power supply capacity for a powerful GPU. Even HCI platforms that use full-size 2U servers usually only have space for one or two GPU cards.
When it comes to hyper-converged infrastructure and VDI, this limited GPU capacity results in limited maximum graphics performance for virtual desktops. A single GPU per host is likely to provide great graphics performance for business applications, but it may not be sufficient for more specialized uses. There are other hypervisor hardware platforms with 4U or 6U servers that allow four or more GPUs per server.
Three-node minimum raises costs
Most, but not all, HCI products do not use RAID inside each node but instead use replication between nodes to protect against storage hardware failures.
These HCI platforms require at least two nodes, and often three nodes, before they can run even a single virtual desktop. HCI may not be a great choice for VDI that is deployed to every retail branch, for example, where IT only needs a dozen desktops. The cost of a three-node HCI may outweigh the value of VDI.
A better solution might be to use a stand-alone hypervisor host with only local storage for each branch VDI deployment.
Mixed workloads prove a challenge
Related to the minimum cluster size is the tendency to use the same, minimum-sized HCI cluster for server and desktop virtualization. The challenge is to design the cluster for the combination of VDI workloads that have usage peaks and valleys and a more stable but critical server workload -- while keeping the cost under control.
In larger deployments, IT can usually isolate the server workload from the desktops with different hypervisor clusters and storage. When a minimum-sized HCI cluster has the capacity for the required desktops and servers, IT can often use the one cluster for both. Usually, the challenge with hyper-converged infrastructure and VDI in this case is getting sufficient storage performance without excessive cost.
Both the three-node minimum and mixing workloads on the HCI cluster are more likely to be an issue for remote office/branch office and small business organizations. In an enterprise data center, IT is likely to need far more than a minimum cluster and often multiple clusters.
How to overcome limitations
Several of these limitations are centered on the physical hardware of the HCI platform. A software-only HCI enables organizations to choose the servers that fit their requirements. To make room for more CPU sockets or GPUs, IT might decide to use 4U servers for its HCI.
There are plenty of situations in which HCI is unlikely to limit a VDI deployment. Most HCI nodes can hold more RAM than IT will want to buy for VDI, where CPU is more likely to be the limiting resource. Every HCI platform has options for all-flash storage, which can deliver amazing storage performance.
Even the maximum HCI cluster size shouldn't be an issue for VDI, as IT usually builds VDI clusters with building blocks of between eight and 16 physical hosts. HCI is an excellent platform for VDI, although some deployments may require a specialized implementation to match specialized requirements. Software-only HCI offers the most extensive choice of physical hardware, at the cost of more design time selecting that hardware.