Next Stop for Blades: Virtualization

Carl Weinschenk spoke with Adam Mendoza, senior manager in the Virtualization and Grid Infrastructure Business Unit, NetApp, and a member of the board of Blade.org.

Weinschenk: Where are we in virtualization of blades?
Mendoza: The true measure of the adoption of any technology is whether it is moving into production environments. Quite a number of customers have considered and adopted blade infrastructure in this way. There is quite an increase in blade chassis and blades in them. We now see multiple instances of virtualization in blades. Instead of one-to-one deployments, in which a single server is in a chassis, now they are adding multiple virtual machines. Instead of 14 physical blades in a one-to-one scenario, we have umpteen virtual blades. As you can image, 10 times 14 servers — 140 servers — for the price of 14 is a significant consolidation. There also are significant savings in power, cooling and space as opposed to standalone servers.

Weinschenk: It would seem that the connection to the storage servers also gains in importance, since a virtualized environment means that each connection carries a greater amount of data. Is this true?
Mendoza: From the network storage side, it's a critical element of making solutions enterprise-ready. You have to implement technology that can connect to these blades, so that the data can be resident outside the server. As far as the state of the industry, people have been using blades and experimenting with virtualization and using network storage. Now they can deploy all three. We are right at the cusp of trying it out. [People say] it's interesting, it seems to have significant benefits. Now they are beginning to deploy business- and mission-critical applications on the combination of virtualization technology, blade servers and network storage.

Weinschenk: What else can be gained?
Mendoza: There is another critical factor to that. The economies of this combination of technologies is a huge factor when you look at the macroeconomics of the industry. There is a significant push to do a lot more with a lot less. Though there is an initial deployment investment, it pays back in months or weeks, depending on customer requirements. The traditional way to deploy business apps is via dedicated servers with dedicated storage. The problem is that there are so many of those deployed that companies run out of real estate in the data center. The form factors are not conducive to rack and stack. There are network connectivity and power requirements. The cooling systems also are fairly independent. The need is to bring these things together in terms of a physical consolidation. The challenge was deploying these servers. They need environments that make it easier to manage and share resources, such as power and cooling network systems. The last piece of the puzzle is sharing the storage.

Weinschenk: It seems to offer great advantages.
Mendoza: The ROI is obvious. There are shared cooling systems and power systems. The next piece is shared storage. Companies are starting to save very real dollars. We are beginning to see very real returns on operational costs, on the management of that environment, and on the deployment of new applications and services.

Weinschenk: There is an issue of whether business continuity was robust enough.
Mendoza: Fundamentally, it came down to, "Can I use this technology to failover my business applications?" A lot was happening between blades, chassis and data centers. Time passed until they eventually matured enough. Now people say, "I can put all my eggs in one basket because I can move my eggs anywhere I want."

Weinschenk: How is the physical plant improved?
Mendoza: First, let's look at micro efficiencies, if you will. There are efficiencies in aggregating 10 cooling fans into one cooling system. Another aspect is the reduction in power distribution units — PDUs — that aggregate power systems into the blade chassis. Then I look at peripheral kinds of things like network interfaces. Fibre channel has a bigger power draw than Ethernet cards, but [the gain of using Ethernet is not so much] if you have one Ethernet card in each chassis and separate power for each of these interfaces. With blades, you can use consolidated power supplies. That part of it is very easy to understand. It's watts for watts.

Weinschenk: What about the cooling.
Mendoza: I think that aspect is even more critical. If you standardized on blades, you can redesign the data center so that you can have hot and cold aisle implementations or architecture. Essentially, you can start putting blade chassis into racks and having your overall computer room AC push cold air into one aisle and warm into another. Lo and behold, the cold aisle draws it in and pushes it and next has the hot air that is all being drawn back into the AC units. You don't have hot and cold mixing. The temperature differential is usually about 20 to 30 degrees.

Weinschenk: There seems to be a tremendous upside to all this. Do you agree?
Mendoza: The estimates are that only 10 to 11 percent of production servers have been virtualized. If we look at consolidating onto blades, there is still a lot of room for these technologies to be deployed and for things to leap forward. I really do think — and not just because I am a storage guy — that overall efficiencies are going to be driven by storage infrastructure because you will want to rapidly provision your virtual machines and business applications from images available from your network storage. This will make business flexibility a reality. That implies a couple of different things. People will look at the deployment of new business applications and all the infrastructure and operational management elements of that environment. I think we are going to look at the business continuity aspect and all that implies as being a lot more of a norm across all your business applications. In the past, it was too complex. With virtualization on blades and network storage, we can now deploy for a majority if not all applications.