Measurement of power consumption in enterprise data centers and measurement of power consumption in data centers

Source: Internet
Author: User
Tags benchmark

Measurement of power consumption in enterprise data centers and measurement of power consumption in data centers

Currently, there are many ways to allocate, estimate, and measure energy consumption in the data center industry. In this article, we will discuss with readers about the reasons for measuring the power consumption of data centers, the measurement of power consumption, and how to handle the data collected. We will also introduce some new available technologies. technology.

Why do we need to measure the power consumption of the data center?

Assume that you, as the data center manager of your company, receive a call from the CIO asking you: "What do we do in terms of data center power consumption ?" What will you do? Usually, data center managers do not worry about power consumption. After all, after seven o'clock P.M., other fields of an enterprise may run without air conditioners, but data centers will never work. If you need additional IT-based services, You need to purchase and install necessary hardware. This method is changing rapidly because 1) additional power resources are usually unavailable, 2) power costs are becoming an important cost for operational data centers, and 3) Now, more and more enterprises are attaching great importance to the adoption of green environmental protection measures, so that they can be recognized as excellent enterprises by the society, shoulder their social responsibilities, and be prepared to comply with various environmental protection laws and regulations.

Obviously, you cannot effectively manage things that are not measured. This is especially true for energy consumption. In this case, the estimation results made by the empirical law alone are likely to be incorrect, resulting in unnecessary and sometimes considerable cost loss. Devices that are considered to consume only a small amount of energy may consume a lot of energy, even if they are only idle and do not perform any substantive work tasks.

The first step is to set a benchmark for the current power consumption of the enterprise data center. Ideally, this will be compared by providing useful historical statistics. Early measurement and estimation may be rough, but over time, the power deployment inside and outside the data center is gradually better understood, and the measurement quality is gradually improved, therefore, it can be improved over time.

There are many ways to manage the power consumption of data centers. However, without some benchmarking measures, it is difficult to know where to start or which measures can have the greatest impact. In addition, without baseline measurement, it is impossible to display the management of energy consumption levels and the improvement of energy consumption in your enterprise data center.

Efficiency Indicators

Currently, the most important indicator of energy efficiency in the data center industry is PUE ). This is the ratio of all the energy consumed by the Data Center (including IT equipment) to the energy consumed by IT equipment. Total consumption includes the energy consumption of lighting, cooling and airflow management equipment, and power distribution units in the data center. IT devices are the devices that perform computing tasks.

  • PUE = total device energy consumption in the data center/IT device energy consumption

The data center that only supplies power to the IT device will reach PUE = 1.0, because both the numerator and denominator are the power of the IT device. This is obviously not in line with the actual operations of the data center. Even if all the lighting systems in the data center are off, the corresponding power resources will be consumed to provide cooling and air flow management needs, and the distribution efficiency will be low.

The average enterprise data center efficiency (CADE) Benchmark takes into account the energy efficiency, utilization and server utilization of data center facilities.

  • CADE = (facility efficiency) x (IT assets efficiency)
  • Facility efficiency = The amount of energy provided to IT equipment/The amount of energy obtained from public power supply companies

IT asset efficiency = the average utilization rate of all the server's central processor (CPU) (usually a small percentage, such as 5%) until efficiency such as virtualization is implemented.

Where and how to measure

In the data center, power can be measured in multiple locations. From the most rough measurement to the most detailed measurement, the first step is to make a measurement at the position where the power enters the data center. If the data center is an independent structure and relies solely on power supply from public power companies. The measured value is the total power in the PUE formula.

But in many cases, this is not easy. Enterprise data centers may only occupy several floors of a building. In this case, separate electric meters should be installed on the floor or room where the data center is located. If the data center does not share power or the relevant facilities in the building (such as cooling equipment), the electric meter records the total power. If facilities and power are shared (especially in city data centers), data center administrators need to estimate at least the total power consumption of the data center, which may come from several different sources (for example, total power supply measured by electric meters into the data center, plus a certain percentage of power used by building cooling equipment ).

The next location where frequent power measurements are performed is the uninterruptible power supply (UPS ). If IT only supplies power to IT devices, this data can be used as the denominator for PUE computing. However, UPS may also provide power for rack-mounted refrigeration devices.

The third position of power measurement is the rack itself, which features the Distribution Unit (PDU) of the metering rack. These measurements are generally considered to be an IT device integrated into the rack, unless there is a fan or rack-side cooling unit.

The fourth position of the power measurement is the sockets of the PDU in the rack. These smart PDUS usually also provide integrated Rack Power consumption measurement data. The power at the outlet layer is monitored to ensure that the power consumption of IT devices can be identified in PUE computing. Specific measures can be taken to improve efficiency by providing power information at various device levels.

The fifth position of the measured power is on the CPU. This gives a measurement of the power consumed by actual computing. In fact, this is not widely used today. CPU-level measurement is not very useful in taking practical energy-saving actions. In most cases, data center employees can change or reduce data on the entire device, blade server, or other IT equipment, rather than the CPU. The most typical method for measuring data center power consumption is metering rack PDU and smart rack PDU, which are used to monitor a single output.

How to process collected data

Different measures to improve energy efficiency can be taken based on the measurement location and measurement method selected by the enterprise data center. If you can provide useful and operable information, we recommend that enterprise data center managers use independent outlet-Level Measurement Methods for IT devices.

By monitoring the power consumption on the rack, data center managers can determine whether the original power distribution is reasonable. Generally, IT equipment is allocated power based on the nameplate rating, but these nameplate ratings are usually very conservative. Even if a certain percentage of nameplate power is used, for example, 70%, the power is usually excessively allocated. This means that the power consumption of the IT device rack will exceed the actual power consumption. This "Idle Power" can be deployed elsewhere, but how can we know that the racks in your data center are not likely to suffer from depletion of power resources during load peaks?

Regularly monitor each device, and the shorter the interval, the better, to ensure that the peak periods are not ignored. With the power consumption data of a single device, you can set a rack to supplement the power consumption mode of the device, so that you can use the same power to support more IT devices. If the rack is about to consume all the power resources allocated to IT, and thus there is a risk of interfering with circuit breakers, having independent IT device power consumption data allows IT administrators to remove the device in a reasonable way, this minimizes the risk of circuit breaker trip while maintaining the appropriate load.

For example, by testing in its own data center, the US company Raritan determines the simple empirical rule that the nameplate rating percentage does not work. Of the 59 servers, the average power consumption of 15 servers is 20% or lower, 29 servers are 21% to 40%, 9 servers are 41% to 60%, 4 servers are 61% to 80%, and 2 servers are 81% or more. Even at peak power consumption, the nameplate rating for 49 servers is less than 60%. Many data center planners use 70% of the nameplate, which means many data centers have a lot of idle power.

In terms of peak power consumption, 5 out of 59 servers account for 81% or more of the nameplate, which may be disabled. In terms of power consumption, it is very important to understand what is happening on a single device, rather than simply understanding the overall average values that may mask high-end and low-end problems.

Environmental sensors and Their Impact on power and cooling efficiency

Environmental sensors play an important role in improving the power supply efficiency of data centers. It is not uncommon for cooling to consume 30% of the total power of more data centers. The supplier provides the inlet temperature specifications. As long as the inlet temperature is within the specification range, the server can work normally. These specifications are generally much higher than the specifications normally provided in the cold channel of the data center. Therefore, you can increase the operating environment temperature of the data center to reduce the power consumption of the cooling device.

The temperature sensor should be placed at the bottom of the rack on the air-conditioning inlet side, 1/3 in the middle and top. Cooling IT equipment below the required temperature will consume a large amount of power resources without any adverse impact. Due to the lack of rack-mounted instruments, data center managers often experience excessive cooling to ensure that IT equipment will not become invalid.

Introduction to available new technologies

It is not enough to take a single power snapshot at a time point. IT devices may consume much less energy at eight o'clock A.M. and may reach the peak power consumption at four o'clock P.M. on Thursday. Power Consumption may also change with the seasonal changes in the year, for example, the peak of online sales in December.

Some hardware devices can take a power snapshot every several seconds at user-defined intervals. The software program can be used to convert these data points to the calculation of power consumption. The measurement unit is kWh ). More advanced tools can calculate carbon footprint based on energy usage. Based on the actual information of a single device, data center staff can learn about the units that generate the largest carbon emissions, so they can manage them as shown in the following figure.

Related considerations

Accuracy: due to carbon emissions ceiling, credit and transaction mechanisms are adopted, making accuracy important. Assuming the perfect sine wave (rarely seen in the real world), the accuracy of plus or minus 5% deviation may be acceptable to determine whether the rack runs at a margin of about 25% before the breaker trip. This is unacceptable when laws and carbon credits are processed so that they can be verified and traded. The billing or chargeback refund is not accurate enough.

Openness and interoperability: IT management systems are deployed in many data centers. To associate such a system with power measurement, open standards for integration and interoperability with existing devices are needed. Ease of use is an important consideration, so power management is not a time-consuming project for busy IT staff.

Security: power resources are the lifeline of data centers. It is important to ensure secure access to the power management system. Search for systems with high-level encryption, such as the 256-bit Advanced Encryption Standard (AES) and the ability to set authentication, authorization, and permissions.

Conclusion

We hope that if your company's CIO calls you next time and asks, "What do we do in terms of data center power consumption ?" You can refer to the content described in this article and outline a feasible plan to establish a baseline by collecting relevant data information. Now, data collection and adoption of data center indicators such as PUE computing will help your enterprise data center manage power and power costs more effectively. In this way, you can more confidently answer the phone number of the CIO.

Contact Us

The content source of this page is from Internet, which doesn't represent Alibaba Cloud's opinion; products and services mentioned on that page don't have any relationship with Alibaba Cloud. If the content of the page makes you feel confusing, please write us an email, we will handle the problem within 5 days after receiving your email.

If you find any instances of plagiarism from the community, please send an email to: info-contact@alibabacloud.com and provide relevant evidence. A staff member will contact you within 5 working days.

A Free Trial That Lets You Build Big!

Start building with 50+ products and up to 12 months usage for Elastic Compute Service

  • Sales Support

    1 on 1 presale consultation

  • After-Sales Support

    24/7 Technical Support 6 Free Tickets per Quarter Faster Response

  • Alibaba Cloud offers highly flexible support services tailored to meet your exact needs.