Data center software and hardware design from the perspective of Energy Conservation (1)

Data center software and hardware design from the perspective of Energy Conservation (1)

-PMC Senior Consultant, former Facebook Storage Architecture designer,

Per Brasher, one of the founders of OCP, recorded in the  speech-

The speech process is as follows. First, we will discuss the reasons for focusing on the efficiency of the data center and the original motivation for Improving the efficiency. Step 2: Discuss the main components that affect efficiency, the contribution of these components to efficiency, and how to optimize and improve the efficiency of each component. The third aspect is how to further reduce TCO costs, which will involve more advanced data protection mechanisms. Next, we will make a certain outlook on the various new storage modes, and finally make a summary.

The following table summarizes the OCP design philosophy, which lists each component of the data center and its weight in the total cost, the measures taken to optimize the design concept of each component and the proportional impact on equipment investment and O & M costs are also listed. The conclusion is that, based on the design concept of OCP, the weighted average cost of equipment investment is reduced by 23%, and the O & M cost is reduced by 33%.

It looks complex and shows the environmental factors in data center operations, mainly temperature and humidity. Traditional data centers usually have a narrow environment, strict requirements, low temperature and humidity. Such an environment requires a large investment in air conditioning cooling, from both equipment and air conditioning and cooling energy consumption. At this stage, we are aware that by easing the working conditions and easing them to the A2 range, we can greatly improve the energy usage efficiency of the data center. The blue line in the figure shows the range currently supported by OCP. The energy efficiency of pue1.07 is achieved through changes in heat dissipation and air conditioning design. If you are not familiar with PUE, simply put, it refers to the ratio of overall data energy consumption divided by the energy consumption ratio of IT devices. The closer this ratio is to 1, the better.

What does this mean? The increase in energy efficiency is directly reflected in the cost savings. The power consumption per watt is about $1.9 per year. If a megawatt data center drops PUE from 1.9 to 1.07 per megawatt, it means a savings of $ more than 1 million. This figure is impressive for some data centers that consume more than 1 megawatt of energy.

Next we will talk about it from three aspects-from big to small. The first is the design of data center equipment rooms (including air conditioners), which can save energy. The second aspect is the power distribution system, how to distribute electricity, and some design improvements in this aspect. In the third aspect, the rack and server are described in detail.

How to Design air conditioners? External air enters the data center. If the air temperature is too low, the heat of the existing data center will be mixed with it through circulation, and some filters will be used. If the temperature is too high, the solidified device will take away some heat through the gasification of water, and then send the cooling gas to the IT equipment of the data center. After IT equipment, the temperature of the gas increases, or it is recycled, or directly delivered to the outside of the data center.

The left side shows the traditional distribution mode, usually 480 v ac incoming, after AC/DC Conversion and UPS, switch back to AC, and then send it to the rack through peu, there is actually an energy loss in each step of conversion. The total loss is about 27% when it reaches the server. The figure on the right shows that in the OCP design, the 480 V power distribution is directly sent to the rack. The more efficient OCP power supply and 48 V Ds are used for protection, and the total energy consumption is 7.6%, this includes the power supply loss. In the past PUE evaluation, the power loss inside the server is not taken into account. Therefore, the above improvements not only greatly increase the number, but also far exceed the earliest design results.

The three figures below represent the evolution of three generations of OCP servers. The first generation and the second generation of the chassis design is exactly the same, the first generation of the motherboard is very large, only one; the second generation is placed on two motherboard; the third generation has changed the chassis design, however, the motherboard is the same as the second generation. During the evolution process, OCP has a strong awareness of Supply Chain protection and tries to reuse existing designs as much as possible. As a result, suppliers do not have to invest in development and testing costs for every specification change.

Compared with the cost impact of traditional data center design and OCP design. Red represents the traditional design, while green represents the OCP design. As shown in the figure, compared to the two, OCP can save about 33% of the cost. The blue section shows that using new technologies proposed by PMC can further save and improve OCP. It should be noted that, in terms of storage, the further cost reduction space is still very huge.