By: David Tebbutt, Programme Director, Freeform Dynamics Published: 24th October 2007 Copyright Freeform Dynamics © 2007

The data centre with its huge potential to gobble power and generate heat has long been the focus for green computing initiatives. And quite rightly so. The alternative is arrange rooms so that cold air is delivered to where it's needed most, usually by careful room and rack planning. This still leaves the possibility that power is being wasted because each area has to be designed for the worst possible environmental conditions.

Just under a year ago, HP Labs started talking about Dynamic Smart Cooling. The idea was to place a network of temperature sensors on the racks, providing information back to a control centre which, in turn, controlled the cooling fans in real-time. Such an approach was designed to minimise the energy requirements and, therefore, the costs. And as a by-product, of course, reduce the amount of carbon dioxide released to the atmosphere.

All very well, but competitors complained that this was merely an announcement of future intent. A touch of greenwash, perhaps? Except that Hewlett Packard could demonstrate the system working in the labs. Now, though, the company has its first internal data centre implementation running and and has announced its general public availability.

The first site outside the laboratory is in the company's newly-consolidated data centre in Bangalore. This 70,000 square foot centre is already operational, although it has yet to bring up all of the planned 2,500 racks. The company's six new US data centres are scheduled to be ready in early 2008.

All of these data centres are the result of consolidation and this is, "when the power bill hits you in the face," according to John Sontag director of virtualization and datacenter architecture for HP Labs. In Bangalore, the issue is complicated by the fact that the centre has its own diesel-powered generators and a requirement for diesel oil storage. This is because public power supplies cannot yet be relied on in this part of the world.

With diesel oil in India providing a kilowatt-hour for around 25 cents, Sontag estimates an ROI of six months for the Dynamic Smart Cooling system. Where power costs are lower, the payback time will obviously be higher. Perhaps twice as much in California, say. The company estimates power bill savings of up to 40 percent over conventionally cooled datacentres. Once fully optimised, the Bangalore operation (a consolidation of fourteen centres), is expected to save Hewlett Packard 7,500 megawatt-hours annually and reduce carbon emissions by approximately 7,500 tons. (Based on the figures given, that suggests the DSC cost was just under a million dollars. Anyone from HP care to comment?)

The DSC can be retrofitted to existing data centre racks and legacy equipment. It just requires that the fan drives be variable-speed and have industry standard controls. The controller (accessible remotely as a web service) is the smart bit, able to calculate what fans need to run at what speeds to deliver cooling to the right places at all times. Sontag explains that if a modern high density (30kW) rack loses its cooling, it could overheat in 90 seconds. By the time a human has received and reacted to a warning, it would already be too late.

It's certainly a seductive argument. Hewlett Packard believes it's first with such an intelligent adaptive system. Others will, no doubt, follow. Systems that respond to changing needs will remove a lot of risk from data centre planning while at the same time cutting both energy bills and carbon emissions.

I wonder if anyone's working on tying virtualisation and intelligent cooling systems together?

Just a thought.

Reader Comments

We are no longer accepting comments against this item. We suggest contacting the author directly.

26th October 2007: 'Simon Griffiths' said:

This is great smart use of technology. Too often we see a method of spot cooling, which certainly fixes a cooling and reliability issue, but often does not address the issue of total power usage, and power efficiency. CFD is one option to optimise the position of servers in the datacentre, but that is not easy to do when environments can quickly change. Having this type of control will therefore allow adaptation of a well set up data centre to fluctuating loads.

One word of caution. I have not got a good handle on the software, but with fluid flows it is important that a balance is reached. For example, increasing the flow of a fan here, will effect something in another spot. The system would have to be smart enough to handle this, or it will go into an adjustment loop that will end up defeating the purpose of the system. My advice would be to have a look at the HP experience, and make sure that the system is responding to changes in operation, and not a steady work load.

Reply to Simon Griffiths?

26th October 2007: 'David Tebbutt' said:

Thanks for the comment Simon. My understanding is that the intelligence in the software takes care of the issue you raise.

I should also mention that the mathematics are considerably more complicated than my simple calculation suggests (HP commented about this on the original blog post over at Teblog).

I might also add that HP confirmed to me, following the post, that it is looking at tying software virtualisation with DSC, as I suggested at the end.

Reply to David Tebbutt?

28th October 2007: 'Simon Griffiths' said:

That's good to know, and I would also say that these calculations are very complex, essentially requiring iterative calculations at multiple point in 3D space.

My other concern is that CFD by it's nature can produce incorrect results. This is particularly true when small scale changes effect larger scale flows. Usually someone experienced in fluid flow will cast their eye over results to see if they seem 'reasonable', and adjust the mesh (calculation points) if they do not. This would be extremely difficult for software to achieve, and the only possible solution I can see is using a very fine mesh, which would massively increase calculation times.

Having said that I still think that this is a great way forward, and probably the ideal solution to many power and cooling issues.

Reply to Simon Griffiths?

29th October 2007: 'Bob Sakakeeny' said:

The following is from the HP Labs director responsible for the project: " The CFD analysis of a data center layout is used by the HP Services people to optimize the initial design of an installation. that analysis has a direct benefit on cooling power, as the tools give you a good sense of how to statically provision a data center. We've seen saving up to 25% in cooling power using these sort of techniques. It is also true that a static technique has to plan for the worst case load, and so will tend to over provision the average case, hence the opportunity for Dynamic Smart Cooling. DSC is not recomputing the CFD model of the room, it is using a set of sophisticated control algorithms that monitor the temperature at the racks and air conditioners(CRAC), and computes the optimal fan speed for each CRAC that keeps every system's air intake within temperature limits. Consequently, as systems consume more or les power, the room flow adapts, minimizing energy use."

Reply to Bob Sakakeeny?

29th October 2007: 'Simon Griffiths' said:

It is great that they are using CFD in the initial design, but this approach will potentially give you the problems I was hoping they would avoid by re-evaluating with CFD. The fact that fluid flow is such a dynamic situation will mean that tweaking here and there will unbalance other things in ways that are often unpredictable. Having worked with fluids over many years I know that this is one of the most difficult things to predict. My guess is that the software HP use will have to be learning and will require a lot of calibration for the almost infinite number of possibilities of flows that their system would allow. My worry is that it might not!

Reply to Simon Griffiths?