<b>UPS and Power Distribution design to achieve Redundancy</b>
The article you referenced was excellent and easy to read and took the issue head on with many concepts. Let me take an approach that starts with equipment, a server with 2 power supplies and work my back towards the UPS. Unlike that article, I will not include UPS technical design although I do have to mention capacity.
My first and last high level design concept is that most Tier 2 and above Data Center Power distribution models include the concept of supplying power from two sources, either of which having the capacity to run all of the critical equipment. In this model, let us assume there are two similar UPS systems (i.e. “Symmetrical”) and we will call one UPS A and the other UPS B.
Each Rack would have two Rack PDUs (i.e. “Power Distribution Unit”) , which are similar to power strips except they may be networkable, remotely manageable, and even carry 208 volts for more efficient delivery of power. One of the Rack PDUs would be connected to UPS A and the other to UPS B. Of course, the equipment, servers, disk enclosures, switches, etc., must also have two or more power supplies, with separate power supplies connected to the two sides. For example, a server with two power supplies would have one power supply connected to the “A” side Rack PDU and the other to the “B” side Rack PDU. A detail requirement is that this equipment must be able to operate on half of its power supplies, either the half fed from UPS A or the half fed from UPS B.
Larger servers, tape libraries or communications equipment would have one or more of its power supplies connected to one UPS and the other redundant side connected to the other UPS. (Note: If the equipment power supplies cannot run on one side independent of the other, this is a weak link but does not invalidate the design. Also, some cost saving methods could include connecting the A side to UPS and Generator and the B side to street power – penny wise, and may be a budgetary necessity, acceptable for non critical equipment, but also may be pound foolish.)
In summary, the Data Center, at a minimum would have two UPS systems with one feeding one Rack PDU and the other feeding the Second Rack PDU in each server or communications rack. Large equipment may have its power supplies connected directly to the UPS A and UPS B.
There are more high-level considerations to this such as Generator Capacity, start up load surge, and Air Conditioner loads, which must be supplied during a power outage. At the detail level, capacity planning must ensure that each phase on a Rack PDU, each breaker in the UPS PDU, and the UPS itself on either side, is not overloaded by a failure of the other side. When this is achieved, you have UPS redundancy.