Abstract: In fact, for energy saving, traditional technology is also made “twelve” efforts. However, in the case of continuous technological evolution, the traditional energy-saving technology still has problems, how to break?

This article is from the huawei Cloud community “Data center energy saving? Try Huawei NAIE data center energy saving Technology!” , original author: Qiming.

One, 3 years of electricity consumption, can build another data center!

1.1 Driven by science and technology, data center market continues to develop rapidly

Data center: A data center is a globally coordinated network of specific devices used to transmit, accelerate, display, compute, and store data information over the Internet network infrastructure. The primary purpose of a data center is to run applications that process data from businesses and operational organizations.

Today, we are in a fully connected world. From 2015 to 2025, according to huawei GIV data forecast, the number of intelligent terminal connections in the world will surge from 7 billion to 40 billion, and the number of global connections will also surge from 20 billion to 100 billion. Behind the surge in hardware and connections is an explosion in data traffic: annual data traffic will surge 20 times from 9ZB to 180ZB (see figure 1).

Figure 1: Data from HW GIV

The rapid growth of data traffic, coupled with the government’s strong support for various emerging industries, data center development and construction will usher in a period of rapid development, according to the statistics of MarketsAndMarkets estimates, The value of global data centers will grow from $13.07 billion in 2017 to $46.05 billion in 2022 (see Figure 2), with a Compound Annual GrowthRate (CAGR) of 28.9%. Its market size and market value are self-evident.

Figure 2: Data from MarketsAndMarkets

1.2 High power consumption, “shadow behind” data center industry

“There is always a shadow behind the sun.” Behind high industrial value is high electricity consumption. As a “data center”, you can imagine: a large computer room, which is densely covered with various cabinets, servers and so on. The upfront infrastructure and investment in data centers will be huge. And once start to use, this electricity charge among them, again will be an astronomical figure. We can look at the power usage of a large data center over 10 years of operating costs:

As can be seen from the table above, the data center uses nearly $36 million in electricity per year, 70% of which is used for electricity, and 19% of which is used for cooling. And according to 2017 statistics, global data center electricity consumption accounts for 3% of global electricity consumption, with an annual growth rate of more than 6%, equivalent to 30 nuclear power plants. Data centres in China alone use 120 billion kilowatt-hours of electricity a year, more than the three Gorges power Plant produced in all of 2017 (100 billion kilowatt-hours). If you calculate the electricity cost of a data center for 3 years, you can build another data center!

1.3 External policies + operational challenges, energy saving in data center industry becomes an inevitable trend

The data on electricity bills behind data centers are so shocking that there are national policies that have strict requirements for energy efficiency. For example, the Ministry of Industry and Information Technology’s Green Data Center Guidelines require new data centers to have a PUE of less than 1.4. Beijing, Shanghai, Shenzhen and other cities also have planning requirements for PUE, especially shenzhen Development and Reform Commission encourages the PUE of new data centers to be less than 1.25, which is actually a very challenging figure. Of course, the European Union and the United States have their own regulations for PUE. Energy conservation, after all, means lower costs, which in turn increases profits.

To solve the energy consumption problem, we need to first list the energy consumption problem into a formula, and then reduce or increase a certain value of the formula to achieve the purpose of reducing the energy consumption. This formula is the calculation method of PUE that we talked about earlier.

PUE stands for Power UsageEffectiveness. PUE= Total energy consumption of the data center/ENERGY consumption of IT devices. Total energy consumption of the data center includes energy consumption of IT devices, cooling and power distribution systems. The value of PUE must be greater than 1. For example, if PUE=2, that means that for every watt consumed by an IT device, an additional watt needs to be consumed to distribute and cool IT. Of course, in an ideal world, if all the power is consumed by IT equipment, that is, all the power is used for production, then the PUE is equal to 1.

The following figure shows details of energy consumption units in a data center:

As you can see, the energy consumption units of a data center include chillers, pumps, IT equipment, fans, fresh air lighting, etc. The energy consumption of these units is in the molecular position. The closer PUE is to 1, the higher the efficiency is, the more electricity and money are saved. So to save power, IT’s natural to start with the molecular, non-IT energy consumption (mainly refrigeration).

1.4 Find the principle, how to cool the data center

Before coming up with solutions, let’s take a look at how data center cooling works (below is a schematic diagram of cooling).

The whole system can be divided into two parts: refrigerating station and terminal machine room. On the left side of the dotted line is the refrigerating station, which includes cooling tower, refrigerating unit, water pump with various functions and cold storage tank for cold water storage. To the right of the dotted line is our IT equipment room, in which there is an air conditioner to blow out cold air in addition to the server cabinet. The cold source of air conditioning comes from the refrigeration station on the left.

To put IT simply, the whole system refrigeration system is to move the heat emitted by the server in IT equipment to the outdoor. The power consumption units of the refrigeration system are also very intuitive, such as cooling tower, cooling pump, cooler and air conditioning on the picture.

Of course, the diagram above is just a simple schematic, and the actual refrigeration diagram would be far more complicated. So how do we save energy in complex systems?

1.5 Limitations of traditional energy-saving technologies in the evolution of technology

In fact, for energy saving, traditional technology is also made “twelve” efforts. However, with the continuous evolution of technology, traditional energy saving technologies still have the following problems:

  • The application of product-grade energy saving technology has reached the ceiling;

  • The system is complex, there are many equipment, and the influence of energy consumption among equipment is complicated, which is difficult to be simulated by the traditional engineering formula. The traditional control methods are independent, and the effect of expert experience has reached its limit.

  • Each data center is a unique environment and architecture, and while many engineering practices and rules of thumb can be applied across the board, a customized model of how one system operates does not guarantee the success of another.

NAIE Data center energy saving technology how to help energy saving

2.1 Industry consensus: AI helps save energy in data centers

As mentioned above, traditional energy saving technologies can no longer meet the demand of data center energy saving. People began to look for new ways.

Today, the industry consensus is to use AI to regulate the entire refrigeration system, matching the performance of each piece of equipment to achieve optimal performance. According to Gartner’s user research, 30% of data centers that are not ai-ready will no longer be economically viable by 2020. The survey also identified three ways ai could improve data center operations:

  • Use predictive analysis to optimize workload allocation, implement optimized storage and calculate load balancing;

  • Machine learning algorithms optimise transactions and artificial intelligence optimise data centre power consumption;

  • Ai can ease personnel shortages and automate system updates and security patches.

“Using AI to Regulate refrigeration systems”, most notably Jim Gao and the DeepMind team. They used a neural network to predict PUE, data center temperature, and load pressure, respectively, to control variables for about 120 data centers and achieve a reduction in PUE.

The industry has been using AI technology for data center energy saving with great success. Let’s take a look at how NAIE data centers are helping to save energy.

2.2 Huawei NAIE Data center energy saving technology

In terms of “energy saving”, in fact, is a very big topic, and NAIE data center energy saving, also includes many aspects, our introduction today, to “refrigeration system energy saving”. For “cooling system energy saving”, NAIE data center energy saving has the following four “means” :

2.2.1 Original data feature engineering

Data center cooling systems typically have complex piping layouts, installed refrigeration units (pumps, towers, etc.), and numerous sensors in addition to these devices. In addition, different data centers have different locations, resulting in different pipes and devices.

In view of these data differences, we can shield them through AI algorithm. We can deal with some complex structures through feature engineering, such as single pipe, mother pipe and circular pipe. According to different controls, we try to extract unified features, and then extract relatively similar features comprehensively for different equipment, such as cold towers, chillers, heat switches, water pumps, air conditioners, etc. Finally, the data is checked, the missing data is supplemented, the wrong data machine is corrected, and the abnormal samples are deleted.

Therefore, through feature engineering, we can process the data collected at the site into a relatively uniform form and provide it to the subsequent AI algorithm.

2.2.2 Energy consumption prediction and safety guarantee model

To save energy, we first need an energy consumption prediction model. Establishing a good model is a good beginning to predict how to regulate the energy saving of refrigeration system. But there is one big difference between a predictive model for industrial control and a model for predicting stock movements or subway traffic: safety control. After all, safety production is the first, saving electricity and money is the second.

So the NAIE Data center Energy prediction model is not a simple, stand-alone model, but a set of models that not only predict energy consumption after adjustment, but also predict the state of individual smart systems. To ensure the normal status of all systems on the basis of energy saving.

2.2.3 Optimization of control parameters

The introduction of the first two “means” has laid a good foundation for energy-saving algorithms. In the third “means”, will be “results”. Whether or not the control parameter we search for is “excellent” is entirely determined by the quality of the third “means”. The “Energy prediction and Safety Assurance Model” provides a good model for energy consumption and state prediction, which can be thought of as a hypersurface graph (as shown below). Of course, its shape is impossible to draw and difficult to imagine, because we are dealing with a problem of high dimensional space, and there are many holes in the hypersurface, which represent insecure control parameters. So the purpose of our third “means” is to quickly and effectively find the best or optimal control parameters and send them to the equipment for execution.

2.2.4 NAIE Cloud-ground collaboration

Cloud-ground collaboration is an automated service that integrates the cloud and the ground to realize the whole process of data collection on the cloud, model daily evaluation, chong training and model update.

Just to be clear: data collection, new samples; Daily evaluation of the model, i.e. deciding when to update it; Retraining, that is, the process of retraining, finally achieves the goal of full automation of model updating. (See below for concrete frame drawing)

NAIE’s cloud-to-ground collaboration, which includes NAIE’s data Lakes, data center PUE optimization model generation services, and the AI marketplace, which manages generated model packages; At the local end of the client network, there is the network AI framework (the platform that runs the model generated service generated models). The local network AI framework is responsible for sample collection and management, as well as continuously evaluating the generated model with new samples. If it is found that the distribution of collected samples has obvious changes, or the model accuracy is always below the standard, it will be triggered to rebuild the model.

In addition, the network AI framework connects to the actual control system of the DATA center through Huawei Cloud Opera Neteco. In this way, the control parameters generated by the model can be directly delivered to the actual group control system.

2.3NAIE helps save energy in data centers

In a huawei data center, with the help of NAIE, the PUE of the whole year was reduced by 0.12 compared with that before using AI, which translates to 328.6 kilowatts of power consumption per sampling cycle. That works out to 5.8 million yuan in electricity savings a year, a significant figure.

  • NAIE model generation service

Different data centers, in the refrigeration mode (water cooling, air cooling, AHU, etc.), pipeline type (mother tube, single tube, mixed tube) and other aspects are likely to differ, how should we start with it?

This is where “feature engineering” comes in. As we mentioned earlier, feature engineering is useful for masking the many differences in AI algorithms and trying to form uniform features.

General modeling (as shown in the figure below) for developers: from energy-saving modeling to model application, it takes 4 developers six months.

NAIE, with the help of “feature engineering” and “old expert” technology, has prepared the preorder conditions for you. Let’s take a look at several highlights and advantages of NAIE:

1. Zero coding efficient modeling: Model of data center topology templates, AI based on huawei training platform and PUE features/algorithms library, energy engineers only need to provide infrastructure operation data and refrigeration equipment technology parameters without any code, can be matched online AI model of its data center, model development time reduced to 1 person from June 8 * * 1 month, The development investment of the whole model was reduced by more than 95%;

2. Flexible and visible parameter configuration: Based on huawei’s visualized parameter configuration in the DC field, PUE models of DCS in different topology templates can be generated by adjusting parameters.

3. Comprehensive control strategy: by importing PUE related full parameters of data center infrastructure, the model can deduce the control strategy of a full set of refrigeration equipment, such as chiller, cooling pump, cooling tower, refrigeration pump, plate exchange, etc., to help energy engineers flexibly and accurately regulate the refrigeration system to achieve the best energy consumption state;

4. Good optimization effect: Through professional feature recognition and processing, the model fitting effect is good. Under the premise of quantity and quality assurance, PUE prediction accuracy reaches 95%.

Through the data center PUE optimization model to generate service website (console.huaweicloud.com/naie/produc…). , you can quickly experience the service: Click “Function Demonstration” :

Go to the service introduction page and follow the instructions step by step to experience the PUE optimization model generation service in a fast and convenient way.

Data center PUE optimization model generation service combines AI technology and data center engineering experience to provide automated modeling tools (such as data center topology template, PUE feature/algorithm library, model training platform) to help data center field engineers with 0 basic 0 coding. They only need to input the operation data of data center infrastructure. You can get an effective PUE optimization model online.

Click to follow, the first time to learn about Huawei cloud fresh technology ~