This article can also be found in the Premium Editorial Download "Storage magazine: Status report: Solid-state storage."
Download it now to read this article plus other related content.
A systems integrator on the front lines of virtual desktop infrastructure (VDI) implementation offers some solid advice.
The following is a discussion I had with Keith Norbie, vice president of sales at Nexus Information Systems. Nexus is a system integrator that focuses on virtualization and storage. Keith and I talked about virtual desktop infrastructure (VDI) and its implementation challenges directly related to storage.
Tony: What’s your role in your organization and what have you been doing so far in the VDI space specific to storage?
Keith: I’ve been with Nexus for 12 years and in the industry for 20 with a focus on storage in the last 11 years.
We’ve developed a storage vision with virtualization as the centerpiece. We’ve sized workloads for databases or backup, and looked at I/O profiles in a number of different characteristic formats and scientifically measured them. IOPS is an important factor for VDI to deliver what I call a “DirecTV” system for users.
Tony: You take a look at how many desktops you’re going to have and what the I/O workload is going to be on a storage system.
Keith: Yes, and there’s another depth beyond basic IOPS analysis. There are characteristics, including how bursty and unpredictable those IOPS can be. We get into the scientific measurement process to understand how users are using workloads. Everyone’s got a laptop or some kind of I/O initiator in each of their
Tony: It happens frequently.
Keith: It puts your computer into like a 30-minute freeze. Imagine that across 1,000 units. You just completely took your entire back office down. That doesn’t happen pervasively across every single user using Word at the same time, but it’s the things you haven’t really experienced yet in the modern massive implementation of VDI that become factors you have to account for.
Tony: I was just talking to a school district that has approximately 1,000 users on VDI right now. They were getting boot storms eight times a day because students log in and log off with each period.
Keith: I’ll give you another real-world scenario. I ran across an organization that had a SAN. They had plenty of storage with lots of free capacity. Everything was running. They almost felt their storage was idle. They naturally just threw in the VDI and virtualized some desktops for users in their call centers. Well, the problem was the I/O generated on the SAN had 15 spindles and they couldn’t generate enough IOPS for their couple of hundred users. They simply created a very bad user experience. They couldn’t figure out that it was because they simply were running it on an under-horsepowered array.
Tony: Have you found that the initial adoption of VDI has been stalled because of not really understanding the nuanced issues created by virtual desktops?
Keith: Yes. I’ll just say the technology available from multiple array manufacturers and VDI software makers, and the technology that’s out there, is very robust and very doable. But it’s both a lot harder than you think and a lot easier. Folks who want to just jump off and do it and not think -- those are the folks who are going to get burned in this area.
Tony: What are some of the ways you can overcome the IOPS issues within a VDI environment?
Keith: It depends on the environment and how much, in each of those areas we discussed, you need to factor for. In the case of the IOPS requirements, just throwing more spindles at it impacts heating, cooling and rack space. Yes, you can just throw in more spindles, but that comes with a significant downside. You need to factor in the front end plus managing all that capacity. It’s a dynamic math equation that’s really individualized per environment. I’m not a fan of ripping out everything and bringing in something totally new. I want to understand each of the components in the environment and then modify to meet the mathematical fulfillments of what the environment is telling you.
Tony: And you can build these mathematical models even around the lack of predictability?
Keith: Yes, because a great deal is predictable. Our measurements give us all the known usage spikes, whether it’s from boot storms, antivirus or other factors. We know when people are logging on, but every environment is a little different.
Tony: Planning some performance headroom for those things you can’t predict is very important.
Keith: Exactly. And there’s another factor; there are a lot of emerging technologies. Taking advantage of some technology breakthroughs can be really beneficial to get them that plus one.
BIO: Tony Asaro is senior analyst and founder of Voices of IT.
This was first published in October 2011