The conversation in Track 3 on Day 2 was very focused on whether reasonable and relevant metrics can be constructed that work across industries and companies. In fact, one speaker has a model for seven different types of data centers that he thought would deserve their own metric - so that if you are in that category, you can assess your internal strategy and milestones with others.
A basic premise of determining the efficiency of a data center is the Power Usage Effectiveness (PUE). Your theoretical goal is to achieve 1:1: that is one kilowatt of power to produce one kilowatt of compute power. (“We’re really close”, whispered one participant to me – “we think 1.5 is high”). However, in most data centers the PUE might be between 3 and 4, and the wisdom here is that it takes 4 - 7 kilowatts to produce one kilowatt of compute power from “chip to grid”.
But even with agreed-upon metrics, you could arrive at a similar number based on vastly different designs – some with raised floors, some with other efficiency measures in place. Setting a target in your overall design and strategy, based on expected load, will give you a number to aim for. Then the changes aren’t linear, and you will be able to predict with more accuracy.
Here are random notes based on the conversations around me.
The real business issues…
1. Cost and availability of energy – and coal is still the cheapest across America. That is what is driving the move of data centers out of California and toward San Antonio and Austin, Texas. Increasingly climate zone has an impact because free “outside air”, as it is sanctioned by ASHRAE can bring down the annual energy bill.
When a state such as Wyoming was interested in attracting new data center business, they offered to provide a minimum of 10% renewable energy while the rest of the existing energy is from coal.
2. Access to fiber optic networks
Depending on where you locate, the cost to tie-into the network costs can be hundreds of thousands dollars. You might have to build your own fiber optics. And you will probably prefer redundant fiber optics networks.
3. Ongoing cost of operations.Staffing is not a large percentage of budget as more tasks have become automated. For example, a 450K sq foot data center may be managed by just 100 people and a 100K sq foot data center may require only 30 people.
4. Classification of your data center (Tier 3 or Tier 4) makes a difference in what PUE your data center can achieve. The higher the tier, the more difficult it is to reach a 1:1 ratio because of the higher reliability requirement.
5. The electrical mechanical infrastructure is the most costly aspect. This is where the mechanical tracks at the workshop spent most of their discussion. It seems that ASHRAE and NEBS, one association for the facilities industry, and the other for the computer industry, have different recommended and allowable recommendations which are confusing to data center managers. The work assignment from this charrette is to reconcile those different recommendations so that implementation is less of a "guesswork" exercise. The work that needs to be done is benchmarking all the variations between the recommended tolerance levels so that the risk level is easily understood.
6. The maturity of the company and the industry it is in result in different “hot buttons” for executive management. Where cost savings are important in mature industries, cost of opportunity is the metric for emerging industries.
7. Facilities engineering culture: Looking at set points on CRAC units is difficult for many of the facilities engineers that most companies employ. If the skill set isn’t inhouse, it has to be accomplished through a $200,000 consulting contract and it is difficult to predict the savings in advance to argue for that level of investment.
Real estate (land and building shell) typically comprise 5 – 15% of the total budget and the differential is by location.
What’s green about all of the above? Higher efficiency means a more productive use of energy. If your company is growing, then IT has to grow as well. But turning off unused appliances and powering down underused hardware hasn’t been done in tandem and everyone agrees with the statement that waste is everywhere. Andrew Fanara of the EPA addressed the group in the evening and stressed the message that electricity will be more costly, regardless of whether it is renewable or not, so building energy efficiency into every data center's strategy is an imperative.
It was a great workshop and demonstrated the need to bring the two industries together, facilities and IT, to work on problems that plague both of them in day to day operations. Defining the problems in terms so that benchmarking and recommendations can be made is the next step, and the participants agreed to work on those over the next six - eight months until their next workshop in 2008.
Have a green day!