3/2 N: how to save on improving reliability?

Parallel configurations of uninterruptible power supplies (UPS) have been used in the industry for more than 35 years to protect critical loads. They are also used to ensure the fail-safe operation of critical data center applications. But which backup scheme to choose when building a commercial data center in conditions when, on the one hand, the budget is limited, and on the other hand, the protection of client equipment and applications is the number one priority?


Currently, the most widely distributed parallel systems with N + 1 redundancy (see Fig. 1). If one source fails, the system will remain operational due to the fact that the backup UPS takes over the load. A parallel system can theoretically be assembled from any number of sources - the main thing is that their total power provides the power rating that the load consumes, plus one backup source. In a parallel system designed in more detail, the excess of the total power over the nominal value will be less. For example, if the load is 400 kVA, then you can take three UPS of 200 kVA (200 + 200 + 200). The excess of the power of such a system over the load capacity will be 200 kVA (50% with respect to the load). But if you take less powerful UPSs, for example, 4 UPSs of 100 kVA,

It would seem that the fractional scheme is more economical ... But in practice, taking into account the cables, commissioning, installation, installation, and even just the cost of the UPS, these benefits are not obvious, and the fractional system takes up more space ... Therefore, for all that it is attractive at first glance prefer large fractionality, the N + 1 scheme usually comes down to 2 + 1, 3 + 1 - and extremely rarely to 4 + 1.

In data centers with a critical load, the N + 1 scheme is used extremely rarely, and even then because of the limited budget. The weak point of this circuit is a single point of failure: the bus through which the UPS is connected to the load. If it is necessary to carry out routine maintenance (for example, when it is necessary to tighten the bolts on the bus), the load will have to be disconnected from the UPS group. The most effective strategy in this case is to connect the load to the second feeder from the city.


The ideology of the TIA 942 standard for data centers requires that the system continues to be operational, even if any of its elements fail. Considering that almost the entire modern load has two independent power supplies, it is quite natural that a different connection scheme, “2N”, arises. Each of the inputs (server power supplies) is powered by a separate UPS group (see Fig. 2). Currently, the desire of equipment manufacturers to increase the number of power supplies to three or more has been outlined. So if, after five years, three or four power supplies will be widely practiced, the redundancy scheme may accordingly change from 2N to 3N or even 4N.

The comparison shows that the number of devices in the second scheme is greater, and accordingly it is more expensive. But from the point of view of reliability, the failure of any source or the whole group does not lead to loss of load, which smoothly goes to the backup input.


art of designing power protection systems allows you to maximize the reliability of the system, but at the same time significantly reduce capital costs. Such a solution is the scheme shown in Figure 3. Of course, this scheme is not an innovative discovery. In the world of DRUPS - diesel rotary UPS systems - this circuit is called IP-Star.

In Figure 4, the same pattern is presented in the form of a star. Its use in the new DataPro data center in Tver allows you to maintain the level of reliability almost the same as in the 2N scheme, but at the same time significantly reduce capital costs.

Three server rooms will be commissioned in the new data center. The total load consumption power of each room is 400 kVA. Each load in this circuit is connected to two active inputs. Each UPS group includes three sources with a capacity of 200 kVA each - two main and one backup. The total installed capacity in this case is 1800 kVA. The diagram shows that when one of the sources fails, the load is not likely to be lost. And even if an accident occurs on the centralized bus of one of the UPS groups, the neighboring UPS group will pick up the load from another input. Obviously, if there were not three, but four rooms in the data center, then the scheme would be called not 3/2 N, but 4/3 N.
In normal operation, each of the loads is protected by uninterruptible power supplies from both inputs. If one of the rays disappears, operability is maintained. The number of UPSs in this circuit is less, in comparison with the circuit 2N (3x200x4) = 2400) the system is cheaper - the advantages of the third circuit (3x200x3 = 1800 kVA) are obvious.

In general, in the third scheme, all the advantages of the 2N scheme are preserved. In this case, the UPSs are loaded at 2/3 of the nominal, and not at 50% as in the 2N circuit. Therefore, the efficiency is higher - accordingly, the electricity bill is less. True, for high-quality sources, efficiency, depending on the degree of workload, does not degrade very much. But in general, both in capital costs and in operating expenses, scheme 3 is cheaper than scheme 2N, although it is somewhat more expensive than N + 1. And this is with almost the same level of reliability as 2N.

At the DataPro company data center in Tver, the 3/2 N redundancy scheme is the first change in Russia and is certified by the Uptime Institute.

Briefly about the main:
• DataPro Data Center in Tver
• Total data center area of ​​2650 sq.m.
• Maximum power 4.5 MW
• 4 machine rooms with the possibility of placing up to 100 racks each
• Design capacity of IT equipment from 3 kW to 20 kW per rack
• Certification for compliance with Tier III Uptime Institute Design and Facility *
• Commissioning October 2013

Expert commentary:

Sergey Ermakov , Inelt Technical Director.
The problems that arise in the second connection scheme mainly lie in the plane of interaction with IT services. In fact, during the operation of such systems, the absence of one input for the operational team of the IT department means an alarm: the signal comes from the monitoring system and is recorded in the event log. It doesn’t matter that the switching was provided by the circuit solution and the process didn’t stop - for IT this is an occasion to write reports that the power engineers did not provide one of the inputs and start an internal investigation.

That is why, where a highly reliable solution is required, redundant UPSs are provided on two shoulders, but then a 2N circuit threatens to transform into 2 (N + 1), which corresponds to Tier 4 and makes the solution even more expensive. There are intermediate solutions - N + (N + 1), when only one of the branches is reserved. The trade-off is that in this case the number of alarms is reduced, but they are not eliminated at all.

To shorten the duration of the alarms, you can enter a large number of cross joints, which in manual, automatic or semi-automatic mode in the event of an accident will allow you to start a backup UPS. But traditionally, specialists are wary of mutual overlapping connections, because when operating such systems, the probability of error by negligence is high. This can also be fought, but in general, all the tricks only complicate the scheme, which ultimately reduces the overall level of reliability.

Yuri Kopylov - Technical Director of Eaton
Currently, data centers have an urgent task - backup power virtualized servers in a cloud computing environment. Depending on the manufacturer, this problem is already being solved by means of the UPS and the corresponding software and monitoring systems, which allow not to wait until the power at the input completely disappears, but transfer critical applications to other virtual servers operating in another zone. Such solutions have already been proposed by Eaton: the UPS and the virtualization system have already learned how to transfer the most critical applications to those virtual servers where there are no problems with the power of the physical IT infrastructure.

Interesting solutions can be obtained if modern “modular” UPSs are used in any of the considered schemes, where each of the UPSs is not a single 200 kVA device (as in the examples), but itself consists of intelligent modules operating in parallel mode. They not only provide internal redundancy and some redundancy of the UPS itself, increasing its reliability, but in conjunction with other UPSs of the parallel system form a certain “matrix structure” that automatically redistributes the total load among the working modules.

Also popular now: