A key management principle is that you can’t manage what you can’t measure. So if we want to manage the energy efficiency of our data center estate, then we had better be in a position to baseline where we are today so that we can identify the value of the changes that we implement.
The problem with measuring anything is deciding what to measure. Measuring data center efficiency is particularly difficult, where do you start and what do you measure? A good starting point is looking at what others do in this space.
A common measure is PUE – Power Usage Effectiveness a simple measure of the total power delivered to the facility divided by the IT Equipment Power. So how, then do we measure or calculate PUE?
To get the Total Power Delivered to the Facility, and if we had instrumentation in our data center, we could measure the instantaneous power entering the building through the high voltage feed (often this feed has other uses including ancillary buildings, offices and other non-data center assets which will need to be excluded). We could alternatively look at an average and use the utility electricity meter to count the number of MW hours over a fixed period (dividing the MW hour value by the number of hours we measure over gets us to an average power value).
To get to the IT Equipment Power value we have two approaches, we could use the boilerplate values (vendor supplied power rating usually attached to the equipment on a label) for each piece of equipment and simply add them up or we could (if we are lucky to have them) use our iPDU devices to measure the total power delivered to the racks. Both of these approaches have weaknesses. In the former case, we will find that boilerplate values are very poor measures of the real power consumed as that will be often dependent on the number of accessories attached or the loading of the device (e.g. a network switch). Getting the values from the PDU is much better but will tend to be too large as often rack mounted fans and other non IT Equipment load will be hanging off the PDU feeds.
An alternative simple measure to PUE is DCE – Data Center Effectiveness, the reciprocal of PUE i.e. IT Equipment Power divided by Total Equipment Power delivered to the facility. (Another term for the same measure is DCiE – Data Center infrastructure Effectiveness).
The main weakness with these measures is that they focus exclusively on the building M&E effectiveness and fail to address efficiency in terms of computing, network and storage. In other words a good PUE does not mean that you have an effective data center if the servers are all running at 2% utilization and the storage is mostly empty. Nevertheless highly efficient M&E plant is a great place to start.
A fuller measure that is sensitive to more of the initiatives we can take to being energy efficient is CPE – Compute Power Efficiency. This is IT Equipment Utilization divided by PUE.
A facility with IT Equipment running at 5% utilization will have a much lower CPE than another with a better PUE but a 15% IT Equipment Utilization. This is a much more realistic measure of what has been achieved. Plainly equipment running at 100% utilization with a PUE of 1 is the target that we will never get to but nevertheless it should be an aspiration.
Even CPE is a rather one dimensional measure as it focusses on Servers, there are many other areas of efficiency that we could bring into the equation:
- Network Utilization
- Storage Utilization
- Data Center Utilisation
Where we need to get to is a meaningful measure that is possible to calculate repeatedly (better still in real time) that we can use to compare our performance with other Data Center operators.
Is it too fanciful to imagine the CPE measure being adopted by government to regulate data center operators in an energy constrained world?