The problem of the effectiveness of small data centers or "why super-modern DCs Google and Facebook do not solve anything"

Most of the discussions about the inefficiency of data centers as well as their “narrow” places are relatively inadequate server load. The point is that in most data centers the equipment is far from being used 100%, most of the servers are working with incomplete loading, and some are completely idle. According to experts from Stanford, about 10 million servers worldwide are idling, which costs a huge amount of $ 30 billion annually. Major losses - wasted electricity.

But there is another problem. This is an inefficient cooling. It's no secret that cooling can eat up about 50% of the energy consumed by the entire data center. Of course, DCs of such giants as Google, Facebook, Microsoft work very efficiently, and the energy here is spent rationally. They write a lot about these companies and their data centers, they talk a lot about them. But all this is just the tip of the iceberg. The rest of the iceberg is DCs of unknown companies, small data centers, which are mass. Nobody writes about them and very few people know, but they are the main energy consumers in the IT sector.

At the same time, very inefficient cooling systems often work in such DCs. In most cases, because the owner simply does not have the funds (and the desire) to upgrade the infrastructure. In the case, if the DC is part of a state organization or a research institute, the team, in general, doesn’t care how much energy is spent - everything is covered from above, and the DC’s leadership may not think about expenses.
')
This year, the company Future Resource Engineering, engaged in the design and support of DC, found a way to improve energy efficiency in just 40 DC. Savings in this case amounted to 24 million kWh. Basically, the team has been improving cooling. The area of ​​these data centers ranged from several hundred to several thousand square meters.

And the main problem of these DCs (as well as many others) is excessive cooling. However, most owners of data centers are aware that cooling is redundant, but there is simply no other way out. There are two reasons for this - back-up cooling system, which can work simultaneously with the main one, and “hot spots”. In the second case, the cooling center has to be cooled more than if all the equipment were heated evenly. Over-cooling is a common, albeit extremely inefficient way to deal with hot spots. For small data centers, this problem is particularly relevant. If some servers heat up more than others, the operator cools the room so that the temperature of the hottest section falls to the required level. And this often means that all other parts of the DC almost frozen.

Another common problem is the weak separation of hot and cold air streams. If the control system air flow is imperfect, then the hot air can be mixed with cold. Naturally, in this case, the cooling efficiency drops, and the cold air has to be cooled “with reserve” in order to lower the temperature of the turbine room to an acceptable level.

There is also a problem when cold air is sucked into the cooling system along with hot air, instead of being used for its intended purpose.

Google, for example, simply does not have such a problem, because every watt of energy is effectively used in the data centers of this company. Manages all this system with elements of artificial intelligence. A similar situation is observed in the data centers of Facebook and Microsoft.

Basically, the priority of the support of a small data center is the uninterrupted operation of its object. How this is ensured — whether by doubling the main systems, by redundant cooling or by something else, few people care. But if the system is reliable and inefficient, in this case the DC owner spends much more money than he would have spent on creating a normal, modern infrastructure management system for his data center. And if nothing changes, most of the data centers will remain ineffective. And no matter how advanced the DC of the telecommunications giants will be in this case - this will not affect the “total temperature in the hospital”.