Data Center Journal

Volume 30 | February 2014

Issue link: http://cp.revolio.com/i/257227

Contents of this Issue

Navigation

Page 26 of 32

24 | THE DATA CENTER JOURNAL www.datacenterjournal.com no way of measuring the carbon output corresponding to its data center's power usage. CUE measures kilograms of carbon dioxide per kilowatt hour of energy con- sumed; again, a CUE of 0 is ideal. Because alternative energy sources like solar and wind create no emissions when operat- ing (equipment manufacturing may be a different story), a data center that relies on them will have a lower CUE than an equivalent-power data center relying on coal energy. is metric may have some value to data center managers looking to maintain a "green" image in the public eye, but it has little benefit in the facility proper. WUE and similar metrics are more valuable in that they measure efficiency of a resource consumed (and paid for) by the company. Changes in WUE, for instance, could indicate issues with the cooling system, particularly if the data center infrastructure and workload remains fairly constant. MeasuRing peRfoRManCe A data center with a great PUE and WUE can still be inefficient. Ultimately, the best measure of inefficiency must take performance into account: how much work is done for all the resources used? Measuring performance, however, is a little more complicated than simply measuring power or water usage. A server could, for instance, be loaded with busywork just for the sake of increasing utilization, but doing so would clearly be unprofitable. An ideal efficiency metric would also take into account dollars earned, but doing so all but eliminates any possibility of real-time measurements. To simplify matters, it's oen safe to assume that the work being done by the IT equipment has some value to the company; the precise value is less important moment by moment. One performance metric is simply utilization. A server running near 100% of its work capacity is fully utilized; pre- sumably, in this case, it is delivering the maximum results. Unfortunately for many data centers, server utilization is low— on the order of 10% in some cases. In other words, these facilities are operating nowhere near their capacity. is fact has driven the trend toward virtualization and consolidation to reduce capital expenses as well as the operating costs of idle (or mostly idle) equipment. Measurements can involve CPU utilization, I/O utiliza- tion or memory utilization. In practical workloads, one is usually a bottleneck for the others, so maximization of all simulta- neously is difficult if not impossible. Another performance metric that is more absolute than the relative measure of utilization is compute units per second (CUPS). A compute unit might, for instance, be flops (floating-point opera- tions), but whatever the unit, this metric represents a certain amount of work done by the IT equipment. Using CUPS provides a greater context than just power consumption: specifically, a metric like CUPS per watt indicates how much work is done per unit of energy (which can also be easily converted into dollars). Adding CUPS information to, say, PUE provides a better picture of data center efficiency than PUE alone, since it takes actual work into consideration. us, when a company improves the efficiency of its IT equip- ment, the poorer PUE value is balanced (or outweighed) by the greater CUPS/watt value. taking Costs into aCCount Although costs are not necessar- ily the first thought (directly, at least) of data center managers, controlling them is obviously critical to running a profitable business. As a result, including costs in data center metrics can be beneficial. For instance, perhaps adjusting the cooling system to consume a little less water but a little more energy improves CUPS per dol- lar. Factoring operating expenses into such metrics is naturally easier than factoring in capital expenses, apart, perhaps, from estimates based on approximate lifespans for equipment. ConClusions Popular data center metrics like pow- er usage effectiveness (PUE) are just the tip of a giant iceberg of possibilities. e key to any metric is measurement of the necessary parameters: particularly, power usage, water usage and performance. Costs may also factor in, although they can be more difficult owing to varying invento- ries of equipment, staff changes and so on. Sticking to just the physical compo- nents (watts, gallons of water, CUPS and such), however, even measurement can be a challenge. But assuming the goal is measurement for the purpose of efficiency improvement, identification of potential problems, troubleshooting and so on, the minor details—such as whether lighting should be part of a PUE calculation or whether power consumption is measured at each individual server—are incidental. Soware and sensors to manage the data are critical. Data center managers simply don't have the time to go around to each piece of equipment and take readings; furthermore, this approach would mean the data could be stale come analysis time. us, data center monitoring or data cen- ter infrastructure management soware can be helpful in collecting information for various metrics, processing that infor- mation and presenting it in an appealing form. Some metrics may be calculated automatically; others may require some extra effort, depending on the soware's capabilities. Although PUE may be valu- able, a world of other metrics exists to aid data center managers in seeing under the hood of their facilities. n Corroded contacts, improper crimping, smaller gauge, or non-copper conductors can significantly degrade your network performance. So if you spent thousands of dollars building a data center, why go cheap and buy patch cords from an online supermarket? Not smart! Cablesys does not sell supermarket grade patch cords. We manufacture high performance, high speed, data center grade patch cords for decades. Cablesys offers 11 colors, 20 lengths, CAT5e, CAT6 and CAT6A for 10G applications. Our mission is to make the strongest patch cord possible so our integrators do not need to worry about one single connection. We also label and bundle patch cords to help integrators save time and money. We then back it up with a 15 Year Performance Warranty. Best of all, we are 40 percent less than most name brands. Give us a try, and you will be surprised how easy and cost effective we are. PATCH CORDS The Weakest Link of Network Cabling We make patch cords the strongest link Call us now at 800.555.7176 to enhance your data center cabling. Free * $20 coffee card...cablesys.com/coffee With $500 order of copper patch cords. See details * . Proprietary design UL ® safety rated Plenum available 1 Assembled in the USA 2 Save 40% or more © Copyright 2014, Cablesys 1. Plenum is only available with solid cable. 2. Available upon request. mpr-0464 - DATA CENTER JOURNAL Februray 2014.indd 1 1/31/2014 8:42:10 AM

Articles in this issue

Links on this page

Archives of this issue

view archives of Data Center Journal - Volume 30 | February 2014