I don’t hide the fact that I’m a huge fan of appliances and that I’m opposed to traditional desktop virtualization architectures. I know all about pod/block design and have designed many solutions using the pod/block principle, but just because I have to use something doesn’t mean I want to. Block designs can be used to scale to the tens of thousands but they are very expensive. The top reason I don’t like this architecture is that you are basically buying a bucket of IOPS, the larger the purchase the bigger the bucket, but no matter what that bucket will fill up, then you need to buy another big bucket. For larger systems these are seven figure purchases and I just find the entire premise complete nonsense.
Pod/Block is a design principle where you buy storage, compute and networking, then mirror it until you have enough of it to cover your user count. The problem is you tend to scale in 500 or 1000 chunks. The appliance approach is completely different. First off, appliance embrace storage and compute happening in the same box. Second, it scales at much smaller chunks 100-200 users. This approach means I’m not buying a big bucket of IOPS for 1000 users, I’m utilzing local storage which tends to be faster (running on the same bus) cheaper (don’t need expensive SAN HDs) and scales better (every time I add an appliance I’m getting more IOPS). The appliance approach in one sense is still pod/block as its still storage/compute/network but it all happens within the singular appliance. The big catch though is that when you add an additional appliance you aren’t adding another separate bucket, you are increasing the size of your original bucket. This means every time you add an appliance EVERY user is affected positively.
Not all appliances are created equal. I will admit I haven’t talked to every appliance vendor out there, but after looking at a lot of appliances one vendor continues to stand out: Nutanix. Nutanix uses a mix of local storage SATA/SDD/FusionIO to give a massive amount of IOPS on each node. This should remove IOPS as being the bottleneck in desktop virtualization projects. Best of all when you add a node, you increase the total size of the infrastructure, e.g. its really a new way of building a SAN. This approach allows me to use local storage for persistent desktops. Most of the time when you are talking about local storage the conversation has to be about non-persistent desktops because if a node goes down you may lose important data. Not true with the grid approach Nutanix makes. For you storage admins think of each node as being an additional controller, if you lose one its no big deal, your data is safe. With all that fluff its only fair to point out that Nutanix is still too expensive, is limited in how they fail to a 1GB link (Update: as of 12/3/12 Nutanix now supports dual 10g), and although I have talked to one customer using Nutanix for 1000 desktops, I still have concerns as how far that solution can scale. So Nutanix isn’t perfect, I just think they are on the right track.
So I really like appliances and based on the “Rapid Desktop Program” announcement from VMware last week it appears that VMware likes appliances too, or so you would think. I have to admit when I saw the Rapid Desktop program I got excited, then I did the unthinkable, I actually looked into the program. To understand what the Rapid Desktop program is here is a blurb from the press release:
View “Rapid Desktop” program gains momentum – Today, Cisco, Dell, HP, and Intel join the growing eco-system of OEM, technology and channel partners that are helping to develop and sell VMware View appliances. These easy-to-deploy VDI appliances host between 25 and 500 virtual desktops that can be provisioned quickly, making them ideal for small and midsize businesses (SMBs) and scalable for larger organizations.
Again that release speaks to me, it sounds like OEMs are building appliances, woo hoo! The problem is this program doesn’t make any sense. I was handed sheet after sheet of every vendor that was in the appliance program and I took every free minute to stop by each booth and get the elevator pitch. What I learned was that to some vendors these “appliances” are just a reference architecture, to another it was a sticker they could put on their SAN and add some reference architecture, to others it was a way to promote a competitor to vBlock, and finally there was Pivot3 and Nutanix (the two solutions I felt where fair to be called appliances).
The Rapid Desktop program reminds me of my MCP (Microsoft Certified Professional) certifications. I have an MCP in Exchange Server and an MCP in Windows Server, you could also get an MCP for SharePoint or SQL Server, what do these various MCPs have in common? They are all called MCP and they are all stamped Microsoft, unfortunately outside of that there is nothing else in common. That’s how I feel about the Rapid Desktop Program. Its a way to get a nice VMware stamp of approval but it means nothing. Appliances should be a single device, not a reference architecture, anything outside of that doesn’t fit the bill as far as I’m concerned.
I am happy to see VMware realize that appliances solve a lot of the cost and complexity issues around desktop virtualization. They still have a lot more work to do to make this program more understandable, and hopefully they next step will be to get the software stack to run as an appliance to (Citrix VDI in a box comes to mind). Simple hardware solutions combined with simple software stacks will bring down the cost of desktop virtualization which in turn will make it an easier for organizations to deploy and maintain.