A study recently released by Forrester research (commissioned by Cisco -- PDF) reveals that virtualization has taken hold in production data centers, but that there are still many hurdles to be addressed--especially when it comes to the relationship between abstracted and physical resources. Titled “How Server And Network Virtualization Make Data Centers More Dynamic”, the 14 page study asked 240 randomly selected firms with experience in managing medium to large virtualized server environments about the barriers that keep them from doing more.
The results are quite interesting.
First and foremost, virtualization has clearly been accepted in production data centers. This wasn’t true even two years ago when I was in the field, but it is definitely the case now. As the chart below demonstrates, very few of the companies surveyed have less than 10% of production systems virtualized. Most have between 20% and 50% of their production systems virtualized, which is amazing to me. What is also amazing is the fact that between 40% and 80% of those systems will be virtualized within the next two years.
This is clearly the age of virtualization. But even more than that, it is the age of abstraction. While server virtualization was initially about “fractional ownership” of physical servers, in order to increase utilization and decrease capital costs, this survey makes it crystal clear that enterprises are moving to using virtualization to make the software systems they are running abstract entities distinct from the hardware that hosts them. They are now managing to those abstractions, and not to the physical box.
The data that makes this most clear is the question asking whether the respondents were using live motion, and how they were using it:
What stands out here to me is that motion is far from relegated to scheduled maintenance activities, but a large portion of the respondents are using it for rebalancing and service level automation. To do this, one has to stop being concerned about what specific server a given workload is running on (though class of server is still important), and start managing to the server workload--the VM--itself.
This is exciting to me for reasons I explained on The Wisdom of Clouds last week. The more you manage to the abstraction, the more attractive it is to build a homogeneous physical infrastructure that can handle any of your heterogeneous workloads. The concepts of Unified Fabric and Unified Computing become incredibly important, and end-to-end infrastructure becomes a better prospect than assembly of individual components.
There is a fairly interesting observation about a new role that seems to be becoming explicit in many IT organizations, the Virtual Infrastructure Administrator. However, the table describing the biggest challenges to supporting live motion in production data centers really caught my eye. The short-short version is that there is a lot of work to do to make this function reliable and easy to use across the enterprise.
Note the need for managing network policy heads the list. I have seen in person the pain that trying to architect networks to support virtualization causes. Doug Gourlay is often heard to say that virtualization broke our network architectures, and I agree. Unified Computing solves a huge part of this problem, but cloud internetworking will have to address this between individual data centers as well.
A revolution is taking place today, and I’m willing to be most of you are a part of it. Now that you have your sleek now virtualized computing platform, new inefficiencies are going to come to the forefront. This is where a new generation of IT architecture will change the game once again.
All images are copyright 2009, Forrester Research, Inc.
You can follow James Urquhart on Twitter.