Once upon a time as an admin you weren’t so sure about virtualization. It was yet another alien technology that management or a well-intentioned zealot in your IT organization was pushing on you, bringing a new set of challenges and performance issues. But today it’s a critical IT component in almost all enterprises small and large. You’ve actually come to enjoy the benefits of consolidation, like a huge decrease in the number of components you manage every day. On the horizon, SDN (Software Defined Networking) promises to take this to its logical all-virtualized conclusion. Server virtualization then is a good thing, so what could be so different about virtualizing desktops? As it turns out, almost everything.
At first glance, how different can server and desktop virtualization be? You have OSs running in a hypervisor, with clients accessing desktops via RDP, RemoteFX, HDX , SPICE, VNC, ALP, PCoIP, or RGS. You set up storage, make sure your rack network pipes are fast, then enjoy the same benefits you get with server VMs. There are even some additional benefits you don’t get with server VMs, like less expensive client hardware, a significant reduction in copper and switch gear, improved security and the flexibility to project a desktop infrastructure anywhere in the world.
In many cases the budgetary attractiveness of reducing those traditional desktop costs is a big factor in the most common failure of VDI deployments. It creates a distracting ROI number that managers don’t like to see wiped out by a critical new cost: a significant investment in high performance storage and low-latency networks.
Engineers dash budget dreams
The director of IT can be almost giddy with anticipation for VDI. First he sees an office without desktops, which is especially attractive while he’s waiting to see if tablets or BYOD really do take over. With VDI it doesn’t matter what the client is, and it can change easily from repurposed older workstations, to license free Linux notebooks, to tablets, to atomic powered wrist watches or Google glasses. OK, maybe interacting with a 15 year-old OS/2 inventory app running virtualized on Windows 8 using gestures isn’t going to work well, but most users will have their familiar desktop in a way they like.
Desktops are a huge cost, a decades long burden that IT managers are eager to see evaporate in a puff of virtual smoke. When the team tasked with the VDI project reminds their director that he’s going to need to find $750,000 for a real SAN, it interrupts his daydream. VDI can certainly reap many rewards including overall reduced cost. But the director was imagining himself at a big meeting, revealing a slide entitled “Total VDI Project Savings” with a huge animated number with plenty of commas. You’re instead telling him he’ll still have to explain technical trade-offs to VPs and the miraculous savings uber-slide doesn’t exist.
Latency and cost are inversely proportional
Storage, and in particular low-latency storage is the key to successful VDI adoption. Network engineers don’t have to think about storage too much- they go talk to that guy who’s been in IT forever and guards the CLARiiON like a mother hen. They are also aware of the eternal pressure for IT to skimp on storage, and increase scrutiny when recommending “big iron” for a desktop project. That under-investment is one of the main reasons VDI projects fail. The pilot project with 25 desktops may work just fine, but the roll out to 500 desktops dies on stage in front of everybody and your V.P., who denied your request for a decent SAN.
The network side is a relative bargain; similar to solving what I call the VoIP ain’t POTS problem. We tell users VoIP is the same thing as switched voice, but in practice it tends to deliver lower quality calls that drive help desk tickets. You can tell users VDI is the same as a desktop but latency issues will result in a sluggish UI users immediately notice. They are on their PCs all day, and a poor performing desktop can affect productivity and perhaps even job satisfaction. As a techie, how many times has an employer’s frustrating, clunky enterprise desktop infrastructure confirmed it was time to seek an interesting challenge elsewhere?
Deploying a few strategic 10G modules with fat links to decent wireless APs is considerably less expensive and easier to manage than deploying distributed Gigabit Ethernet LANs. The concentrated nature of VDI networks can provide plenty of bandwidth while delivering a single, easy to monitor and manage conduit for VDI connections. VDI networks will rely on network monitoring systems more than ever, because they concentrate the majority of critical business activities onto a single platform. Though there is less infrastructure to monitor, single outages affect a larger number of users. Ensuring timely resolution of device outages, analyzing application traffic flow, optimizing QoS policies, reliably managing IP addresses, monitoring virtual networks and troubleshooting wireless will be more important than ever.
Meet your VDI enemy: IOPS
IOPS (I/O operations per second) end up being the devil inVDI storage details. Traditional desktops happily thrash their DAS (Direct Attached Storage). But the virtualized desktop’s storage is remote, and just a little latency between the VM and physical storage has a huge impact on IOPS and responsiveness. Even the best network between the VM host and the client can’t overcome a sluggish virtual drive.
You can tune application servers to perform well on a SAN, optimizing reads and writes into big chunks of data that apps like SQL databases accelerate with CPUs and caching. Desktop app developers however have no incentive to optimize against higher latency disk IO. Even if you optimize the network for streamlined video between clients and servers, you’re still at the mercy of storage latency. For example, desktop browsers still buffer video content to disk before playback, and that’s VM storage IO.
Boot storms, patch application and AV scanning can hammer both the storage network and the array controllers. Data de-duplication and thin provisioning can help but won’t make up for fundamental limits on IO. Fibre Channel SAN will help. iSCSI and NAS are options but are less likely to meet your I/O requirements. They’re also likely to require 10 Gb Ethernet, and that can be as expensive as implementing Fibre Channel. Adding caching or accelerators can help, and for iSCSI or NFS, you can partially offset the CPU hit with hardware initiators. Server SSD arrays are also an option with some vendors claiming infinite IOPs up to petabyte capacity. I remain skeptical of that for now, but affordable SSD solutions are slowly becoming available that deliver orders of magnitude improved access and reduced latency.
Terabytes and Petabytes
The final issue of VDI storage planning is the sheer size of the required pool. You spent years slowly building out a huge amount of distributed storage across your desktops. With VDI you’ll have to forklift in a replacement all at once. If a typical PC had 120 Gb of DAS, 1,000 virtual desktops will require 120 terabytes of storage. 120 terabytes of fast enterprise storage. IT will of course be under pressure to skimp here, with the dream that they can keep a close watch on user’s C: drives and aggressively over-provision. Desktop users however live to fill-up drives and unlike physical desktops, it will be the storage admin dealing with a capacity crisis not the desktop support staff.
Good planning can make you a star
Despite the challenges described above, VDI can provide significant savings over a traditional desktop environment, while providing dramatically improved manageability, security and even energy savings. Done correctly, it even simplifies capacity planning, helpdesk costs and provides a level of flexibility impossible with physical desktops. With careful planning, and the discipline not to cut costs up front in the hopes of more budget later, admins can deliver a dependable solution that admins, users and the budget office will all like.
About the Author:
Patrick Hubbard, Technical Product Marketing Manager and Head Geek
Patrick is a Senior Technical Product Marketing Manager and Head Geek at SolarWinds, an IT management software provider based in Austin, Texas. He joined SolarWinds in 2007 and combines 20 years of technical expertise with IT customer perspective to create geeky content that speaks to fellow networking and systems professionals. Patrick’s previous roles have included product management and strategy, technical evangelism, sales engineering and software development in Austin high-tech and Fortune 500 companies.
Excellent, well-written, keep up the good work.
Excellent writeup. Very practical and to the point. Very useful!