Keeping Your Datacenter Cool

Dave Driggers, CTO of Verari, discusses the challenges of power and cooling in today's datacenters.
As leading sellers of datacenter hardware, HP (Quote), IBM and Sun Microsystems (Quote) love to talk about ways they cool datacenters and conserve energy. It makes them look good to customers concerned about curbing energy-related costs.

In the last few years, those vendors have led the resurgence of cooling heat generated by servers and storage arrays with chilled water. Some analysts praise the way HP, IBM and Sun have taken an old concept and put new twists on it.

Verari Systems isn't buying it.

Like HP, IBM and Sun, Verari offers traditional rack-mount servers, storage systems, blade servers and software to manage its systems. Unlike those 800-pound gorillas, Verari is a virtual startup with an annual $100 million run rate.

Yet Verari believes it has a few aces in the deck. It claims to be the first server maker to create a true storage blade, and it has patented a new way of cooling blade servers in the datacenter.

Using a vertical cooling technique made possible by employing perforated tiles under the blade server rack, Verari claims it can provide up to 25 percent more power efficiency and can require up to 50 percent less cooling than blades from IBM, HP or Sun.

The man behind the innovation, Verari CTO and co-founder Dave Driggers, discussed the vertical cooling method and why rack-mount servers will disappear in a few years.

Q: Why are hearing so much about power and cooling efficiencies just in the last four or five years?

In the last five years, we've seen the power consumption in datacenters almost quadruple. A standard rack five years ago was using anywhere from 1 to 3 kilowatts of power. Today, the average is 5 to 7 kilowatts, but anything that is using our blade racks or IBM or HP blades are pushing 20,000 to 25,000 watts of power.

The second thing is the price of power has gone up. It's skyrocketed. The cost of power as a percentage of the IT spending has tripled in the last five years. We've gone from being under a percent to almost a 5 percent of the IT spend. Some guys believe the cost of power for a datacenter will become 10 percent of the IT budget. IT budgets aren't getting bigger yet power's going up and labor continues to go up. They're not spending any more on the hardware so you've got to find a way to get that budget back by either dropping the cost of the management and power. I don't see power ever getting cheaper.

With the initiatives coming from the president's office and the Senate, the EPA is going to be responsible for knowing how much power is in the datacenter and ultimately what that power is being used for because power soon is going to be allocated to these guys. There is a limited amount of power. Being able to say, "This application is using this much power," we believe is going to be an important play in the future. Most people aren't really out of space. They're out of power.

Q: What is Verari doing to alleviate power concerns?

When we built BladeRack 2, the goal was to build and design it from the datacenter in, as opposed to from the rack in.

BladeRack 1 was designed to look at the rack. How can we cool a rack better? With BladeRack 2, we said we need to figure out how to cool and power a datacenter better. We built a subsystem on the BladeRack 2 to support up to 480 volts. The bulk of the [uninterruptible power sources] in this country put out 480. It's then stepped down to 208 or 110. When you step down, you immediately start losing efficiency. As you distribute the power, the lower the voltage, the more is lost in the actual transmission. So you want to keep them as high voltage as possible.

We're able to deliver 480 volts to the rack, which saves at the UPS. We then take the 480 and convert it to DC (direct current) because DC is better for distribution within the rack. We convert to high-voltage DC in the rack, distribute high-voltage DC and convert at the blade level to usable low voltages. So we virtually eliminate the loss of getting the power from the UPS to the actual blade and then do a high-efficiency convert at the blade.

Q: When you look at IBM, Sun and HP, do you think that their cooling methods are lacking?

Absolutely. That's why they are looking at other methods like liquid again. If you look at the people that are cooling specialists, none of them are saying we need to move to liquid. With our vertical cooling technology, we can cool 34,000 watts of power on one rack. Today's current technology does not allow us to provide any more power density than that. We've got probably one of the most densest datacenters built around our gear right now. But they're just limited by the space of the power generation. In order to generate more power, they start using more space. We're able to cool with air.

The future really is whole-room cooling or whole-datacenter cooling. That's what our architecture already does. We've got customers today that have very large datacenters that are supporting 10,000 machines with 20,000+ watts of power per rack.

You walk in the datacenter and it's 80 to 85 degrees in there. There are no cold spots, but there are also no hot spots. The key is utilizing all the air that's in there for cooling. You want to put back in the hottest air, equally distributed, possible so that you can get that full 30 to 33 degree delta. They're capable of cooling 33. but only if the air is already warm. They can't make the air any colder than 55 degrees. You want to be right in the band where you're getting the full 33 degree delta but you're also keeping the systems as cool as what's cost effective.

We perfected the vertical cooling method. We didn't enhance it with BladeRack 2 but with the power subsystem being more efficient, we got the ability to cool more. Instead of cooling 28,000 watts into a rack, we can now cool 34,000 because more of that is being used for computer rather than waste.

Q: How else does Verari differentiate its blade portfolio?

Our blade competition, which is HP and IBM, have lots of blade servers. What they don't have is a full storage offering no blade. HP introduced something they call a storage blade. It's six small disks on a blade that are really direct-attached storage. It's not really a storage blade. The total density on the rack is going to be somewhere in the neighborhood of 17 to 30 terabytes of storage. We're at 360 terabytes today, and in April it jumps up to 548 terabytes.

That's storage on a blade. It's the densest storage in the market. It's still truly blade-based so you're capable of hot-swapping the drives as well as the blades. It's not designed to be a product that you have to take offline in order to change a drive.

The HP system, you can't change a drive. The drives aren't hot-swappable. To me, that's direct-attached storage; it's just adding capacity to their standard blade. It's the same thing IBM does. IBM has one blade and if you want any extra capacity on that server, you take the next slot and put in two hard drives.

What HP's talking about with the c-Class is the evolution of all of their computing to blades. It's going to happen over time. If you look at their roadmap, they'll be bringing other types of servers and workstations onto blades in '08 or '09. We're bringing that now, so we feel we're a few years ahead of them.

This article was first published on To read the full article, click here.

Comment and Contribute


(Maximum characters: 1200). You have characters left.