Vol.9, No.2

Approaches to Implementing Fault Tolerant Power Systems

By Mark Kupferberg,
Vice-president,
Manufacturing & Engineering,
Kepco, Inc., Flushing, NY

How do fault tolerant power systems end up in modern electronic systems?
Often it goes likes this:

Salesman: I can get an order for a hundred systems, but our competition is offering a redundant power system. What can you do?

Engineer: Why would anyone need redundant power supplies? Our system will be replaced by the next generation product long before the power supply fails. Does the customer need 24 hours a day, and 7 day a week availability?

Salesman: No, but the customer thinks redundancy is better and our competition has it. Can't you just add another power supply and an isolation diode?

The typical Engineering response to this type of market driven requirement is to do just as Sales suggests and create redundancy by basically duplicating the original power system. Even in applications where fault tolerance is a necessity, nuclear power generation or air traffic control, and where redundant power is required as part of the original design, often the same basic approach of doubling the number of power modules is used. This approach comes from using the N+X formula to derive the number of devices needed. N represents the number of power modules required to support the load's total power requirements. X represents the number of power modules that can fail without interrupting operations of the load. Since X is often equal to 1, the redundancy formula, N+X , is known as N+1.

A common approach in applying this formula is to develop a power budget for the load. An engineer then specifies a power supply system where a single power module will provide the total load power requirement. Limited by space and budget constraints and other application requirements, the depth of redundancy may then be specified. (N+1 or N+2. . .) Often, just one additional power module is specified to provide the redundancy. For example, if the load has a 300 watt power requirement, the power system would consist of two 300 watt power modules. This configuration may be described as 1+1 using the N+X formula.

A 1+1 redundant power system design yields a workable fault tolerant design, but there are other approaches that designers may want to consider. A 1+1 design inherently requires that you buy twice as many watts as the system power requirements. If the load's power budget changes, there is limited flexibility (without redesigning the power system) to use modules at a different wattage level. The ability to do field upgrades to a 1+1 redundancy power system may also be limited by the need to change wiring and re-mount components. Also, a 1+1 formulation limits a given power system's use across different products with different load power requirements.

An alternative approach is to apply the N+1 formula with an N greater than one. This implies using multiple smaller power modules to configure the power system. Tandem Computer (now a division of Compaq Computer) used this approach years ago in designing and marketing fault tolerant computers. This approach was a cornerstone of Tandem's phenomenal growth in the late 1970's and 1980's. Tandem realized that as the value of N went up, the cost of redundancy went down as the percentage of capacity that needed to be duplicated dropped from 100 percent to 10 percent or less. Tandem successfully showed its customers that by using smaller Central Processing Units (CPUs) as building blocks, it could provide fault tolerant systems for much less than its competitors because it could use CPUs capable of fewer MIPS to achieve the desired result. (MIPS are Millions of Instructions Per Second.)

The same logic applies to the configuration of fault tolerant power systems. If, to achieve redundancy you double the power, you also double the cost of the power. The bigger the size of the power module, the more pronounced this effect. Compare the N>1 approach for a power system supporting a 300 watt load to the N=1 approach illustrated in table 1. Using the N=1 approach, the power system would have two 300 watt power modules with total system power of 600 watts. With N=2, the system would have three 150 watt power modules with total system power of 450 watts. Using Kepco's HSF line of hot swappable redundant power modules as a basis for doing the cost comparison, the N=2 configuration would cost about 14% less than the N=1 approach.

Table 1

	Value of N
	N=1	N=2 >
Load (Watts)	300W	300W
Size of individual modules	300W	150W
Number of power modules	2	3
Total system power	600W	450W

While the direct cost savings associated with using the N=2 design are significant, the other benefits associated with it may be even more important. First, because of the modular character of the design and the relatively small increment in the size of the power module, the same basic power system could be used for both conventional and fault tolerant configurations. This allows the system builder to have one basic design, supported by only one type of power module. Having to buy, inventory and support only a single power module has important overhead cost implications. The economies derived from buying a single size module in higher volumes also tend to drive down acquisition costs.

Second, responsiveness to customer requirements is increased. Because the power system can be configured to customer order from standard power modules, the lead time associated with procuring the power system can be reduced or eliminated. In this example, two power modules could be used for a non-redundant configuration. Adding a third module would provide fault tolerance. If the end customer wanted a greater depth of redundancy a fourth module may be added.

Third is increased flexibility. In a world where customers increasingly want products that are tailored to their requirements, the need to iterate existing core designs significantly affects the marketability of a company's products. Using larger numbers of smaller power modules to create tailored systems makes this practical. In the above example, if the load requirements grow 50% to 450 watts, all that is required is to slide in an additional 150 watt module.

Fourth is the ability to generate additional revenue through field upgrade of products. The addition of power modules provides increased systems power to support the field add-ons or upgrading. It is simple and cost effective and eliminates the need to gut the original power system.

As compelling as the N>1 approach is with smaller systems, it becomes even more important as the load¹s requirement rise. Beyond, say, 2,000 watts, doubling power to achieve redundancy increases cost significantly because the designer has to buy a lot more power. A load with a 7,000 watt power budget requires the user to buy an additional 7,000 watts using the N=1 approach. The space required to accommodate the additional 7,000 watts of power is also significant. Using an N=7 approach (1,000 watt power modules) like Kepco's HSP modules, adding redundancy only requires buying an additional 1,000 watt module. Imagine the difference between the two approaches if the load's power budget grows to 8,000 watts. The choice dictated by the design approach is to add a 1,000 watt module, find the space and money for an additional 7,000 watt module or design a new power system based on 8,000 watt modules.

Conclusion

There are a number factors that designers need to consider in configuring fault tolerant power system. Using an N>1 approach that uses a larger number of smaller power modules offers possibilities that are worth considering. The N>1 approach offers opportunities in cost, flexibility, responsiveness and revenue enhancements that should be readily apparent from the foregoing.

References:

Kepco Applications Handbook - Glossary
https://www.kepcopower.com/gl.htm

Kepco HSF series
https://www.kepcopower.com/hsf.htm

Kepco HSP series
https://www.kepcopower.com/hsp.htm

Mark Kupferberg is Vice-President of Manufacturing and Engineering for Kepco, Inc. He is responsible for manufacturing operations and design activities. He has been involved in the design and implementation of fault tolerant systems in the electronics, electrical, power generation and process control industries for over 20 years. He is an APICS Certified Fellow in Production and Inventory Management. and holds a degree from Trinity College, Hartford, Connecticut.

The Kepco HSF/HSP series of hot-swappable power supplies provide N+X redundant power in modules sized from 50 watts to 1500 watts. All models contain isolation diodes and circuitry to enable them to share the current into a load. Built in alarms provide a contact closure that opens on failure of any individual module. The front panel of each plug-in power module contains an on-off switch and a "V d-c on" light. When the modules are paralleled, the one with the highest voltage setting automatically assumes the role of "master" and its front panel "master on" lamp illuminates. The other modules become "slaves& and track the voltage setting of the master and share equally in the load's current. Test points are provided to enable the voltage to be precisely trimmed to the load¹s requirement.

HSF and HSP are switch-mode power supplies and incorporate aggressive EMI filtering to reduce the conducted noise below the FCC and VDE 0871 Class B levels.

The 1000 and 1500 watt HSP occupy just a 5" x 5" cross section so that up to three modules will fit in a standard 3U x 19" rack housing. Remote on-off control of the HSP is provided through isolated TTL-level signals powered by an internal 5V supply. Both the output voltage and the current limit are controlled through a 20% to 100% range by an external 0-10V analog signal.

Kepco's fault tolerant plug-in power systems will keep a mission critical system up and running with an extra- ordinarily high level of reliability.