This is the third in a series of blog posts providing tips and tricks for answering common data center questions with the use of SPM. Last month, I gave some guidance on answering, “How do I predict when I will run out of power?” On the other hand, you can gain additional power overhead by improving efficiency.
Question: How do I find candidates for efficiency improvements?
It is easy to say that you should improve efficiency, but how do you know what your efficiency is today and how do you know if that can be improved. The definition of efficiency for any power component or system is simply power out divided by power in (ε=Po/Pi). This leads one to desire measurement of power at each stage of the power distribution chain.
- In the data center, efficiency has to be thought of a little differently. One of the issues with the method for finding efficiency (ε) described above is that the front-end transformer or UPS may be supplying power to more than the IT equipment. In that case, the standard way to talk about efficiency is to use the Power Usage Effectiveness (PUE) which compares the output IT power draw to the total facility power (PUE = PTOTAL/PIT).
- When it comes to the overall power distribution system of a data center, the power out (Po) can be taken as the total power of the IT equipment in the cabinet (PIT) which requires devices such as Server Technology’s POPS CDUs. In SPM, set up reports and trends of Cabinet Device Power to analyze how much power is used by any particular piece of IT equipment and when it peaks.
- For the data center power distribution system, the power in (PTOTAL) can be taken as the total power input of the first device in the chain. This is generally the transformer or UPS. SPM has an optional feature called Custom Device Template (CDT) which allows the user to input SNMP MIB OIDs into the system to directly poll those upstream devices.
- With the data available from (a) and (b) above, one can calculate the PUE of the system. For advanced analysis, some organizations are taking this a step further by identifying how much of the IT power is actually being used for business applications vs how much is wasted while running idle. This is often referred to as Data Center energy Productivity (DCeP) and is beyond the scope of SPM.
- With a method of polling distribution equipment for power usage, the question then becomes one of where inefficiencies occur.
- Most simple distribution equipment will have very low power losses, but active components such as on-board monitoring and communication circuits, any transformation circuits, and bad connections will cause power output to be less than the power input. Suspicious distribution equipment with measurement circuits can often be monitored using SNMP and the SPM-CDT feature. Then comparisons can be made with expected efficiency values for that particular equipment.
- When it comes down to the IT equipment, the manufacturers often provide efficiency data and expected power draw based on the design specifications. The Server Technology POPS CDU coupled with continual monitoring using SPM can provide the data center efficiency or PUE calculation with critical values. By watching trends of power and power factor over time for particular equipment, one can identify potential issues arising within the IT power supplies.
- Overall data center efficiency, especially in the form of PUE or DCeP, depends heavily on usage of the particular equipment. Often, power supplies for servers, network gear, and storage will run more efficiently when loaded more heavily. And, more importantly, unused equipment wastes power in idle states. Here again the Server Technology POPS CDU and SPM can help provide critical data in terms of power trending and a report on low usage power supplies.
- Once measurement data is collected for various points in the data center, the question of what can be changed must be asked. Armed with these answers, an ROI can be built to determine which course of action to take.
- Much of the efficiency analysis hinges on the design and operation of specific pieces of equipment. Identification of low performing transformation or distribution devices often begs the question of upgrading or replacing those devices. A direct analysis of ROI can be done by comparing the cost of energy used by the old devices to the expected reduced cost of the new equipment in question. As seen in (1) and (2) above, SPM is built to provide the bulk of power and energy data.
- PUE and DCeP can be improved by using existing equipment at a higher rate. This is typically done through a consolidation and virtualization project which removes the lowest performing equipment and moves application load to higher performing equipment. Use SPM to measure the effects of any changes done and to prove out the concepts.
- On shared IT resources, especially virtualized servers, the timing of application usage is often the wild card in maximizing PUE and DCeP. Using the equipment 24/7 at 90%+ is a great goal where batch processing is done during non-working hours. This is typically not seen, but one might be able to schedule certain equipment to be powered off during non-peak hours, thus boosting the live equipment utilization toward the goal. SPM can help with this through scheduling of outlet control and trending of device power usage.
- Another data center efficiency, PUE, or DCeP improvement comes from increasing temperature. In other words, reducing the power used to cool the IT equipment reduces the overall cost of running the data center. SPM helps with this by monitoring the CDU attached probes and can compare this data with power usage data in a trend format.
Freeing up lost capacity by improving efficiency can be a great challenge in the data center. For more information on methods for maximum efficiency, read our White Papers on the subject.