SMOKE TESTS

The results of this test were published in many fire and safety journals and resulted in rethinking of fire detection systems for computer and telecommunication rooms. Although these tests were undertaken many years ago, the principles behind them are still valid. We do urge our readers to consider applying smoke tests to their computer and telecommunications rooms. Where these are fire-protected by inert gas drench, we recommend a pressure test to ensure the gas cannot escape quickly when dumped. Typically these rooms may only be 60%-70% airtight, allowing gas to seep out before the fire is extinguished.

Introduction

Fire detection standards for computer rooms are laid down in British Standard 6266 (1982), 'Fire Protection for Electronic Data processing Installations'.  This standard recommends detectors at 25 square metre intervals or "where justified", 15 square metre intervals.  In tape stores, it is recommended that romm detection be fitted from 10 Square metres to 30 square metres. Detectors may be ionisation (heat sensing) or smoke (particle sensing).  BS 6266 suggests a mix of both will improve detection capability, although this is not a requirement. This standard is aged and currently under review: however, some 60,000 UK computer installations have installed fire detection based on its recommendations.

Detector sensitivity requirements are specified in British Standard 5445 Part 7 and need not concern us, except to note that sensitivity testing in turbulent air conditions is not required.

Look up at the ceiling in most large computer rooms and you will see rows of evenly spaced fire detector heads. How did they get there?

In most cases, it was quite simply a sub-contract job done at the time the computer room was built, with minimal regard for what hardware would be installed.  Even if original hardware was considered, several equipment changes may have taken place without rebalancing fire detection.               

Although detectors may be periodically tested, the test usually involves putting a lit joss stick or similar close to the detector heads.  Very few installations carry out tests at floor level.     

The Need for Full Smoke Tests

Air conditioning and fans from hardware create considerable air turbulence.  If you walk round most large computer rooms, you will feel hot spots, draughts, and cold spots.  Tests done at Harwell using computational fluid dynamics, heat transfer and particle trace techniques on the CRAY-2 supercomputer helped to identify the fire spread pattern at the Kings Cross disaster and showed that flames and smoke tend to cling to surfaces and avoid corners - in some cases, flames and smoke may not rise to ceiling level until relatively late after a fire has started.

All these factors tended to cast doubt on the ability of standard fire detectors to detect fire early enough in a normal computing environment.  It was therefore decided to undertake smoke tests in three different computer rooms, all equipped with fire detection to BS 6266. The tests were conducted by Chubb Fire using a standard PVC cable burn-out test and Rosco non-toxic smoke machine (since smoke detectors detect particles, the effect should be the same using a Rosco as for normal smoke).

A VESDA (Very Early Smoke Detection Apparatus) unit was used as a quality benchmark. The tests were recorded on video.

First Test

The first test was in a room housing a VAXcluster with some 5Gb of DASD and standalone VAX minis. The room was about 80 square metres and was protected by ionisation and smoke detectors which had been tested and found fully operational. Air conditioning was bottom-up through the floor void.

The cable burn-out was picked up by the VESDA in less than a minute, but did not trigger the other detectors.  Rosco smoke was continuously pumped into the room and picked up by the VESDA in less than a minute but did not trigger the other detectors - although the test was allowed to continue for some 45 minutes. 

On the video, air turbulence can clearly be seen dispersing the smoke before it reaches the detectors in the ceiling.

Second Test

The second test was conducted in a room of about 50 square metres housing static telecommunications equipment and small DEC hardware.  There was little air turbulence.  The cable burn triggered the VESDA in less than a minute but did not trigger the other detectors.  The smoke test triggered the VESDA in less than one minute and triggered the other smoke detectors in slightly less than three minutes.

Third Test

The third test took place in a 500 square metre IBM room equipped with some 35 conventional detector heads at ceiling level, together with underfloor detectors, and air conditioned top-down.

This room housed two IBM mainframes, some 400 Gb of on-line storage, tape decks and a variety of telecommunications equipment. 

The cable burn was detected by VESDA in under two minutes but was not detected by other equipment.  Smoke was detected by VESDA within two minutes but, after continuous smoke generation for some 45 minutes, it had still not triggered other detectors.  Smoke directed into the floor void failed to trigger underfloor detectors.

Conclusion

Air turbulence caused by air conditioning and equipment fans can defeat smoke detectors fitted to BS 6266. 

Recommendations

Smoke tests should be undertaken at ground floor and waist height level to identify whether adequate detection is afforded by existing equipment.

Smoke detector standards in computer rooms should be to higher standards than BS 6266 and of equivalent standard to VESDA.

Smoke detection should be inserted in, or adjacent to, air conditioning extracts.

© Andrew Hiles