Changes in the concept of UPS power supply and distribution in data centers

With the development of IT technology, especially the rise of cloud computing, the design concept and configuration method of the power supply system as the basic physical facility are also undergoing significant changes. A series of data center construction cases abroad show the following characteristics: improved system reliability, higher power supply efficiency and increasingly shorter backup time. It can be said that UPS power supply is playing a more pure "guardian" role - using less consumption in exchange for higher reliability. This is first of all the need for the large-scale and intensive development of data centers, and it is also a new requirement for the green, environmentally friendly and energy-saving construction of data centers.

Higher power supply efficiency

With the booming development of cloud computing, the increasing energy consumption of data centers has attracted more and more attention from industry insiders. In 2006, the total power consumption of data centers in the United States was about 61 billion kWh, and the total power cost was about 4.5 billion US dollars; in 2011, the total power consumption of data centers in the United States exceeded 100 billion kWh, and the total power cost was as high as 7.4 billion US dollars. The cost, efficiency and sustainability of data centers have been severely challenged. With the sharp increase in power consumption in data centers, "data center energy saving" has become the development trend of future data center construction. UPS consumes 18% of the power in the data center. However, since the UPS system needs to be online 24 hours a day, improving the UPS power supply efficiency can bring direct power savings. Even a slight improvement in efficiency will save a lot of electricity costs every year. The two factors that affect the efficiency of the UPS system are: the topology of the UPS system itself and the design of the data center power supply and distribution that determines the UPS load factor.

The design structure of the UPS system itself largely determines the efficiency. There are two main structures of UPS on the market: online interactive and online double conversion.

The online interactive UPS connects the rectifier and inverter in parallel with the mains input. This design allows the online interactive UPS to compensate for the mains (overvoltage or undervoltage); at the same time, through the corresponding circuit, it can adjust harmonics, transient fluctuations and other power quality problems. When the mains is unavailable or exceeds the set range, the online interactive UPS isolates the load from the mains through a static switch, and switches to a battery or flywheel or other energy storage device to power the load.

Online Double Conversion UPS System Double conversion UPS system completely isolates the load from the mains. It provides 100% conditioned AC power to the load at all times. Since double conversion UPS converts the mains energy twice during operation – first from AC to DC and then from DC to AC, it is called “double conversion UPS”. In normal operation, even if there is no power interference from the mains, the double conversion UPS system will always provide the load with a processed AC signal.

With a double conversion UPS, the power supply must be rectified from AC to DC and then inverted from DC to AC to ensure a complete sine wave and frequency protection function at the output end, as well as to protect the load from seven types of power interference. This method not only exceeds the power requirements of modern IT equipment, but also consumes a lot of energy.

In comparison, the online interactive UPS has a simpler design and fewer components in the main power path, thus achieving higher power supply efficiency while protecting the load. It can be seen that it has become an indisputable fact that the use of flywheel energy storage UPS system can effectively reduce energy consumption.

Active Power provides flywheel UPS, which adopts online interactive design and can achieve the same power output quality as online double conversion UPS. It mainly adopts high-speed microprocessor, which can sample the mains 400 times in a 20ms cycle and respond quickly to transient problems of power supply. The input circuit design of the flywheel energy storage device for the load is almost the same as that of the online double conversion UPS. In addition to saving energy through UPS architecture, the flywheel energy storage system has lower requirements for environmental conditions (temperature/humidity) than batteries. In addition, the flywheel energy storage system occupies only 1/4 of the area of ​​​​the battery, which greatly reduces the requirements for cooling capacity and cooling space, thereby further saving energy. Taking a 15,000 square foot data center as an example, based on the operating power of IT equipment of 50W per square foot, 6.9 megawatt hours (MWh) of electricity are consumed each year. A 5% improvement in UPS system efficiency would reduce energy consumption by 384,000 kilowatt-hours (kWh) per year, or approximately $38,000 in electricity costs (at $0.10/kWh), with significant savings in cooling loads.

As for the main criticism of online interactive UPS, it is the ability to handle the fluctuation of mains frequency. Since the output frequency is synchronized with the mains, Active Power uses the same microprocessor to solve this problem, allowing only +/- 0.2% mains frequency drift. When the mains exceeds this range, the flywheel will be used to supply power. The result of this is that the power supply quality is effectively guaranteed.

Higher reliability

The increasing importance of data information has made the reliability of data centers increasingly prominent in governments and enterprises. Although energy efficiency is becoming increasingly important for data centers, operation and maintenance personnel are still extremely cautious about any changes that may cause downtime risks. More knowledge about data center power supply reliability can help operation and maintenance personnel reduce concerns while improving data center power supply efficiency.

Since 1996, MTechnology (MTech) has been using the scientific method of Probabilistic Risk Assessment (PRA) to evaluate high-efficiency power equipment. Probabilistic Risk Assessment (PRA) is a collection of formal techniques for evaluating the reliability and effectiveness of complex systems. PRA technology combines known information about the failure rates of simple components with formal mathematical models to make a convincing assessment of system reliability. For most electrical, electronic and mechanical components, there is a reference failure rate data. The PRA calculation method refers to this data and takes into account the interaction of each component in a specific system, and combines them with professional knowledge. Therefore, the failure rate of complex systems can be evaluated before the system is assembled.

MTechnology classifies power outages into two categories:

Long utility power outages of more than 10 seconds: In this case, the AC power is transferred to the generator by commanding the automatic transfer switch to switch into operation, and the generator starts running. Power outages of more than 10 seconds are rare in developed countries. In fact, according to estimates by the Electric Power Research Institute (EPRI), customers are ten times more likely to experience voltage sags than total power outages. In the case of total power outages, no more than 4% last more than 10 seconds.

Fault tree analysis shows that automatic transfer switch failure is the most important cause of backup power system failure. The number of automatic transfer switch failures in use accounts for about 95% of the expected number of failures in the system. The model uses data from the IEEE Gold Book, which has a failure rate of about 10-5 per hour, or a mean time between failures of more than 100,000 hours (more than 11 years), which shows that the entire set of electromechanical components performs well in continuous use. Because the automatic transfer switch is only a single point of failure, its failure rate accounts for most of the system failures, and automatic transfer switch failures almost always cause system failures. Because the automatic transfer switch failure rate, the main switchgear failure rate and the generator failure rate are the main failure rates of the system. The figure below is a simplified single-line flow chart of the Active Power flywheel UPS of the fault tree model.

The data shows that the system containing the Active Power flywheel UPS is almost free of failure. The simulations for short power outages were performed more than 100 times as many times as the simulations for long power outages. Increasing the failure rate of short power outages will have a significant impact on the above results.

Short-term mains power outage of less than 10 seconds: In this case, the energy stored in the UPS is sufficient to support the normal operation of the load before the mains power is restored, so there is no need to consider the issue of transferring power to the generator. In this case, the difference in the core reliability of the two UPS systems will be obvious.

In the event of a short power outage, the stored energy is sufficient to sustain the ride-through time of any power disturbance. Assuming that voltage drops and power outages of less than 10 seconds account for 96% of all voltage drops and power outages, this metric plays an important role in determining the reliability of a single UPS system. Fault tree analysis shows that detectable battery failures and undetectable battery failures account for 90% of all double conversion battery failures. The above theoretical failure rates for double conversion batteries are based on the fact that battery failure rates can be reduced with good maintenance and effective testing. Referring to the MTech research results, experience tells us that batteries are unlikely to be as reliable in actual use as predicted by the model. The most likely failure mode in a double conversion battery UPS system is an undetectable battery failure, and testing battery cells is a difficult task. Even using the most optimistic assumptions, even if the double conversion batteries are tested monthly and a large number of battery failures can be detected), the comparison results still show that the flywheel UPS system has higher reliability. In actual use, most battery failures are undetectable, so the flywheel UPS has the advantage. Some experts pointed out that "the most prone to failure in the UPS system is the battery. The only way to ensure that the battery works reliably is to test it regularly. The battery may look good on the surface, but it is actually on the verge of failure. Most data centers use almost 10 times the amount of maintenance that is actually required. And 'excessive maintenance' just reduces the reliability of the system because human intervention is the least reliable of all factors."

The common parts of the two systems are: a single-line mains feed and a backup generator connected to the UPS system input through a switch cabinet and an automatic transfer switch. The figure below shows a single-line flow diagram of the mains, generator, automatic transfer switch and main switch cabinet connected to the UPS input circuit.

The analysis showed that the double conversion UPS system was seven times more likely to fail than the flywheel UPS system. One advantage of the flywheel UPS system is that the probability of the flywheel energy storage system failing when needed is very small. In other words, if the mains power is interrupted when the flywheel system is operating normally, the flywheel power output is almost guaranteed."

Shorter backup time

UPS configuration with 30 minutes of backup battery was once the industry standard, but now it has been shortened to 15 minutes or less. This is the result of data center owners trying to balance reliability and operation and maintenance costs, and also because the fault tolerance mechanism of the data center will rely more on software and algorithm control, and reduce the reliance on infrastructure hardware redundancy.

With the development of data centers, the idea of ​​shutting down thousands or even tens of thousands of IT devices when the mains power is interrupted is no longer realistic and cannot be accepted by the owners; the power supply standard of data centers requires manual switching between two mains power lines or between mains power and backup generators.

The idea that 15 minutes can give the backup generator set "one more chance to start" is also not valid, because if the generator set cannot start within the first 5 to 6 seconds, then just like a car, it is almost impossible to resume starting within the next 15 minutes; the data center power supply standard requires that the backup generator set must be started and loaded within 15 seconds, and the current generator technology can fully meet this requirement.

More and more customers prefer flywheel UPS because if the generator system cannot start and load normally within 15 seconds, then the operation and maintenance personnel are unlikely to find the fault and repair it immediately within 15 or 30 minutes, and the power outage of the data center will be inevitable! "New data centers are beginning to consider whether 30 minutes of backup time is really needed, and whether a shorter time is acceptable? Based on this, some research institutions predict that a change in UPS selection and configuration will be triggered. UPS technology that was considered non-mainstream or niche in the past will be respected in the future. The trend of modular deployment of data centers will surely promote this change.

For domestic users, the reliability of the mains power supply has reached 3 99s (99.9%). Some important loads are powered by dual mains power. The reliability of the mains power supply can be said to have reached 4 9s, that is, 99.99%. In case of a mains power outage, the reliability of the backup power supply can also reach 3 9s. Technically, it only takes 10 seconds to switch from mains power to backup power supply. This is a public standard. At present, Europe has set this time at 8 seconds. At present, many domestic data centers use a structure of 2N or above, and the use of flywheel UPS can fully meet the application.

Recently, in the Changzhou diesel generator factory, the Active Power flywheel UPS and the diesel engine were tested to prove that this statement is true. The data of the Active Power test equipment CleanSource1000KVA flywheel UPS showed that the flywheel UPS no-load and half-load system switching time were 8.8 seconds and 11.5 seconds respectively; when the flywheel UPS was loaded at 55% and 99%, the flywheel discharge time was 36 seconds and 20 seconds respectively. The tacit cooperation of the flywheel UPS and the diesel engine as a "golden partner" has been unanimously praised by everyone. Its power supply efficiency has also been verified: when the flywheel UPS was loaded at 50%, 70% and fully loaded, the flywheel UPS power supply efficiency was 96%, 97% and 98% respectively. The excellent performance of the flywheel UPS in the test vividly demonstrated the excellent performance and broad development space of the mechanical energy storage UPS technology.