Non-stop swapping - the way to zero downtime

A guaranteed, reliable power supply has become absolutely essential for non-stop computing systems and 'mission critical' applications.

A guaranteed, reliable power supply has become absolutely essential for today's non-stop computing systems and 'mission critical' applications such as industrial automation, power distribution and telecommunications.

The capability of a power system to maintain its output to the load at all times is a vital factor in the selection process.

System downtime is often simply unacceptable these days.

This article discusses the essential factors that must be considered when selecting fault-tolerant, hot-swappable power supplies.

High performance, hot-pluggable power systems such as the new TMN Series from Unipower depend on a high-reliability, multi-task connector for each module.

This connector enables a power system module to be rapidly, easily and safely extracted or inserted into the chassis, while the system maintains full power continuity.

The connector must incorporate the following features: blind mating (float mount), low insertion force; high current AC and DC contacts; low-level signal and logic contacts; and pin sequencing with leading earth contact.

Connector pin sequencing is of prime importance for safe reliable operation.

It is accomplished ingeniously yet simply, by varying the length of the pins.

When a hot-plug module is inserted, the first contact must be a safe Earth connection.

This is followed sequentially by AC input and DC output.

The final connections enable input power to the module and release the inhibit function.

Delivering continuous power to a load without interruption suggests a no-fail power system, or one with an infinite mean time between failures (MTBF).

While such ambitions are not realistically attainable, very low system failure probability (or a very high MTBF) can be achieved.

Labelling a system 'fault-tolerant' suggests a probability of failure acceptable to both the designer and the user.

The designer may conclude that some single failure modes are so unlikely that they can be discounted.

Planning a fail-safe solution for every eventuality may be excessively costly and would impose design and physical constraints the user would not accept.

Low failure probability can be improved by using high reliability, specially screened components, or by significantly de-rating the components.

These techniques may also impose space and cost penalties and will provide only a marginal benefit over using commercial grade components.

The approach which dramatically reduces system failure probability, is to deploy hardware N+n redundancy (e.g N+1, N+2 and so on) where N power modules in parallel carry the load and n modules are redundant.

In a redundant fault-tolerant power system, MTBF is vested in the individual power module, the class of redundancy and the time needed to replace the faulty module.

When failure occurs, the full load is carried by the n modules.

During replacement, failure is based on the probability of any n module failing.

If it is replaced fast enough, system failure probability is very low and MTBF very high - assuming the user has spares available.

To provide a real-world example of system MTBF, let's say that a power module fails, must be extracted, serviced and replaced.

For ease of calculation let us assume that this process takes six weeks, or approximately 1000 hours.

Assume a 2+1 redundant power system in which each module has an MTBF of 100,000 hours.

System MTBF then calculates to 1.75 million hours.

This figure, which represents 200 years of continuous operation, may still be unacceptable in many critical applications as it implies a 0.5% failure probability in one year.

MTBF can be boosted, however, simply by having an extra power module available for rapid replacement.

If replacement time is cut to 10 hours, power system MTBF climbs to 167 million hours.

A reduction to 1 hour replacement time pushes system MTBF to 167 billion hours.

Generally speaking, power supplies can be directly paralleled, with or without a current-sharing control circuit.

Without this facility, however, it is difficult to balance the shared output currents and keep them adjusted with time and temperature.

For optimum performance from a redundant power system, current sharing is desirable.

The use of isolation diodes in power module outputs overcomes potential problems when inserting a new module.

These include preventing short circuits that would pull down outputs below specification, and eliminating transients and arcing.

Usually, a low forward-voltage type, such as a Schottky diode, would be specified.

With the correct level of de-rating, this technique will greatly minimise the possibility of failure.

Reverse voltage and switching stresses rarely occur, but in some applications, the use of an isolation diode 'failure detect' circuit is advisable.

Excessive temperature causes more failures than vibration, humidity and dust in industrial computer systems and accounts for 55% of all electronic equipment failure.

Good thermal management at the system design stage will help to avoid such failures.

Overheating in computer systems occurs partly due to heat being dissipated in smaller, overcrowded enclosures where airflow is restricted by cards, cabling, disk drives, etc.

Fans have a lower MTBF than high-performance electronic circuitry and overall system reliability is enhanced when each hot-swap power module has an integral fan.

Comprehensive monitoring circuitry also plays a fundamental role in redundant power systems.

It is vital that faults are rapidly detected, diagnosed and signalled.

Fault signals which the user may require are: thermal alarm, AC power fail, current share alarm, DC power good, OVP latch alarm, low battery voltage alarm, battery charge fault alarm and battery charge rate alarm.

Back to top