PRACTICAL FIRE PROTECTION IN COMPUTER ROOMS AND DATA CENTRES

Transcription

1 PRACTICAL FIRE PROTECTION IN COMPUTER ROOMS AND DATA CENTRES Walker Fire Protection ABSTRACT Information and Communication Technology (ICT) including hardware, software and ancillary services has changed dramatically over the last 20 years. Similarly, the technology and application of fire protection systems used to protect ICT systems has also seen significant changes in technology. As a result there has been a wide variety of fire protection technologies, systems and strategies adopted to provide fire protection to the ICT infrastructure. While some systems such as high sensitivity smoke detection are commonplace, there is a wide variety of design and installation configurations that are adopted. With the advent of powerful desktop computers in the 80 s & 90 s computer systems were decentralised. Mainframe systems were decommissioned in favour of personal computers on each desk. In the internet age and the advent of cloud computing there has been a shift back to centralised systems to leverage processing power and energy efficiencies. Furthermore technology advances have seen introduction of high density computing in the form of Blade servers and virtual machines, where a single machine may host a number (many) of physical and/or virtual servers. This paper considers various aspects of fire protection in ICT Computer Rooms and Data Processing facilities from an equipment, system and strategic point of view. A critical review of common approaches to fire protection within ICT facilities has been undertaken examining the advantages and disadvantages. The objective of the paper is to provide the reader with a view of good practice approaches to various forms fire protection within ICT facilities from a practical point of view. Fire Australia 2010 Page 1 of 18

2 Fire Australia 2010 Practical Fire Protection in Computer Room and Data Centres BACKGROUND & CURRENT TRENDS GENERAL With the exception of military and government applications, computer systems and data processing facilities have been used in organisations for approximately 50 years. Over this time there have been many changes to trends in the way in which such computer facilities are configured and used. COMPUTER ROOMS When computer systems were first introduced into organisations in the 1960 s and 1970 s, they typically comprised a centrally located mainframe or mini-computer, with one or more connected terminals. This mainframe computer was located within its own dedicated room, primarily for temperature, noise and security control. Then in the late 1980 s and 1990 s the desktop micro PC became ubiquitous, organisational computing systems de-centralised and the computer room within an organisations premises all but disappeared. Now in the internet age most organisations run central computing systems to provide services such as file servers, , remote access, intranet, Wide Area Networks (WAN) and customer portals. Furthermore these services are expected to be available 24/7 both to the organisations employees as well as their customers. This change in focus on organisational computing has seen the resurgence of the dedicated computer room within organisations. Furthermore the modern computer rooms are often more mission critical than the computer rooms of the 1960 s and 70 s because loss of the facility affects almost every part of the organisation. For example telephony PABX systems are now another server, so loss of the IT hub can result in loss of all communications to & from an organisation. DATA CENTRES Traditionally Data Centres were built and operated by large organisations that had more than 1 computer, such as government departments, universities or large corporate companies. In the mid late 1990 s the internet was commercially available with sufficiently high data transfer speeds for it to become attractive for organisations to access their computer facilities in an off-site data facility. Consequently since the mid 1990 s there has been a significant growth in data centres globally. The purpose of a data centre is to be able to centralise and share the technical infrastructure (eg power, cooling, security & Fire Protection) required to operate the computers it houses. In more recent times, the focus has been on providing increased levels of availability. A Data Centre enables highly robust and reliable infrastructure to be provided in a significantly more cost effective way than trying to provide the same level of infrastructure to individual systems. Page 2 of 18

3 Practical Fire Protection in Computer Room and Data Centres Fire Australia 2010 There are a number of emerging trends that are appearing in modern data centres: High Energy Density With many data centres located within population centres space has become a premium, particularly as sites suitable for data centres in existing locations are often limited. Thus there is a trend to increase the energy density of existing data centres to enable more computing power to occupy each square meter of floor area. The average modern data centre typically consumes around 5 kw per rack, much of which is converted to heat. However a recent report by Mitchell[1] identifies some new data centre facilities being designed with racks housing 84 servers and consuming 28 kw per rack. Variable Energy Density - traditional data halls within a data centre are generally quite homogeneous, housing numerous equipment racks containing similar equipment. However modern data centres house a variety of equipment some of which will have moderate energy requirements, and others will have high energy requirements all within the same data hall. Hot Aisle / Cold Aisle & Containment To cater for the high energy densities and variable energy densities computer room air conditioning configurations have changed. The current trend is for hot aisle / cold aisle cooling, where cold air is introduced into the aisle between the racks, the cold aisle, and moves horizontally through the rack (using the equipment fans) to the hot aisle. To further improve efficiency a cold aisle containment enclosure is often constructed by installing a ceiling between the racks and normally closed doors at each end of the aisle between the racks, forcing the cool air through the equipment racks. Energy Efficiency While the above trends are related to energy efficiency, or getting more power out of the same space, there is also a trend to other energy efficiency measures such as free cooling by locating the data centre in cool temperate climates and using minimally conditioned outside air to cool the data centre. In Rack Cooling Some vendors are now offering in rack cooling systems which have liquid based cooling equipment located within the equipment racks, rather than as separate air based CRAC unit(s). Modular Data Centres Another trend that is emerging is the concept of modular data centres, whereby the capacity of the data centre is expanded (or contracted) in discrete modules. This means that non-equipment power usage is generally limited to just the equipment in use rather than the future design capacity. Page 3 of 18

4 Fire Australia 2010 Practical Fire Protection in Computer Room and Data Centres AUSTRALIAN STANDARDS DEVELOPMENT Australian Standard AS2834, originally published in 1985, was last revised in 1995 and has not undergone revision since. The standard is currently under revision by a Working Group and a draft is expected to be published in the next months. While the current and previous version of this standard contained a collection of technical requirements, the revision will be focussed on providing a methodology to prescribe measure and predict the reliability of computer environments throughout the life of the environment [2]. The proposed standard will introduce the concept of resilience which essentially is the ability of the facility to withstand or recover from abnormal conditions, fire being one of those abnormal conditions. It is important to note that resilience is a holistic concept and it is not much use having a highly resilient fire protection system if the electrical supply or cooling system is only moderately resilient. The following diagram 1 shows the proposed structure of the revision to AS2834 AS 2834 ICT system considerations for one environment Scope and application (Chapter 1) Environment classification (Chapter 2) Site selection (Chapter 3) Build structure - architecture (Chapter 4) Power - electrical (Chapter 5) Cooling - mechanical (Chapter 6) Safety & security (Chapter 8) 2 Way data flow Telecommunications (Chapter 7) THE STANDARD SERVICE Control (Chapter 9) Operation, maintenance & training (Chapter 10) 2 Way data flow Figure 1 Data centre design and operational strata diagram for proposed revision to AS Diagram provided courtesy of Alan McCubbin, Chair AS2834 revision committee. Page 4 of 18

5 Practical Fire Protection in Computer Room and Data Centres Fire Australia 2010 SMOKE DETECTION GENERAL Smoke detection is a fundamental part of a data centre fire protection strategy. Given that most data halls and computer rooms are typically unattended, it is critical that a fire is detected early so that appropriate action can be taken. Such action may include Initiation of an automatic suppression system. Notification of on-site or off-site personnel. Automatic notification to Fire Brigade for response. Shut down of power and/or other equipment (eg air handling systems). It is common to assume that automatic smoke detection systems are only useful as fully automatic systems, however there is significant value in using automatic smoke detection as an initiator to a manual investigation and response by trained personnel. RETURN AIR MONITORING Computer Room Air Conditioning (CRAC) systems in modern computer rooms and data centres typically have very high airflow rates to provide effective cooling for the ever increasing energy densities. Because of this smoke from an incipient fire will often be drawn directly to the return air of the CRAC air handling units bypassing detectors which have been installed on the ceiling, often to satisfy building codes and standards. So to achieve early detection to facilitate effective manual intervention, smoke detection located at the return air is very important. Ideally the return air detection should be dedicated to return air and not a shared system with the ceiling, particularly in larger computer rooms and data centres. Where an aspirated smoke detection 2 (ASD) system is used with a single detector and shared pipework for both on-ceiling and return air detection, particular attention needs to made to the distance of the sampling pipe from the return air grille and the orientation of the sampling holes. Inadequate consideration of these issues can result in back pressure in the non-return air pipework effectively making those sampling points ineffective. MULTIPLE ZONES / ROOMS It is not uncommon for multipoint aspirated smoke detection systems to be configured to serve more than one zone, such as sub-floor and in room. Furthermore sometimes aspirated smoke detection systems are configured to serve more than 1 room, particularly where the room size is moderately small and they are adjoining each other. 2 ASD is a technology that uses a single detector (usually very high sensitivity) coupled with a fan (aspirator) which draws air from the space being protected via a network of tubes with sampling holes drilled into it, back to the detector. Page 5 of 18

6 Fire Australia 2010 Practical Fire Protection in Computer Room and Data Centres While this type of approach can be cost effective it can also be problematic, particularly as the size of the room increases. The main issue is in relation to air flow within the sampling pipework due to the different pressures experienced within the room. And while the detector heads can generally be normalised so that they don t indicate any fault, the flow in an individual pipe sector may not be adequate and degrade the performance of the system in terms of effective sensitivity and transport time. Given that early detection is so important, wherever possible it is recommended to have separate detection systems for each physical zone. While costing a little more, this approach improves the response, minimises operational trouble with flow problems and significantly improves the reliability & resilience of the detection system. USE OF MULTIPLE ALARM THRESHOLD LEVELS All modern high sensitivity smoke detection systems 3 have the capability of providing alarm signals at two or more smoke sensitivity thresholds. Often this feature is under-utilised or poorly implemented. Commonly automatic initiation of suppression systems and/or equipment shutdown is initiated by very low alarm thresholds (high sensitivity). This can potentially result in unnecessary interruption to the computer room / data centre operations, impacting on the reliability and resilience of the facility. High sensitivity alarm thresholds should generally only be used for notification so that on-site personnel can respond and hopefully take appropriate action before automatic functions are initiated. In fact the alarm thresholds should be set with manual response in mind; with automatic initiation of suppression systems occurring as a back-up should the manual response not occur quickly enough or be effective. AUTOMATIC NOTIFICATION OF FIRE BRIGADE Further to the discussion on use of multiple alarm thresholds, often the installation contractor will connect the high sensitivity computer room smoke detection system to the main fire indicator panel and the fire brigade will be automatically notified at very high sensitivity thresholds. To avoid unnecessary interruptions to the facility and callouts for the fire brigade it is recommended that brigade call should not occur until the detected smoke obscuration is consistent with that of standard point type detectors (eg around 5% - 8% obs/m). Where both point type detectors and high sensitivity aspirated (or point type) detectors are installed only the standard sensitivity detectors should notify the brigade, unless there is a specific early response required. 3 Either aspirated or point type Page 6 of 18

7 Practical Fire Protection in Computer Room and Data Centres Fire Australia 2010 OUT OF THE BOX DEFAULT SETTINGS Often the installer will install a multi-threshold high sensitivity smoke detection system with it s default out of the box settings. While these settings are generally OK for most applications, they are exactly that general settings, which have not been optimised for the particular situation. Most high sensitivity systems have a range of possible configuration settings, such as: Alarm thresholds for each level Threshold delays cumulative or simultaneous Day/Night settings Air flow thresholds Output / Relay configurations Each of the available settings should be considered as to whether the out of the box default setting is suitable for the situation or should be adjusted. Some of these settings can be estimated at design stage, however should also be verified and adjusted during commissioning, particularly alarm threshold settings. The hot wire smoke test specified originally in BS while designed to test the overall response of the system, is also useful for identifying appropriate alarm thresholds for the early stages of alarm requiring investigation. If necessary, larger smoke pellet testing may be used to confirm and verify an upper alarm threshold used for initiating other actions. DETECTION OUTSIDE THE COMPUTER ROOM / DATA HALL Often significant effort is put into early detection of fire within a computer room or data hall, however sometimes there the focus on fire detection outside the room is lost, particularly in computer rooms located within offices. Detection outside the computer room is also an important consideration as a fire can readily occur within office equipment or general building plant that may significantly impact the computer room / data hall. WET PIPE SPRINKLERS APPLICABILITY Wet pipe sprinklers are a very reliable and effective automatic fire suppression system, particularly at preventing a fire from spreading throughout a facility. However wet pipe sprinkler systems are often excluded from within the data/computer room itself due to fear of accidental operation or leakage resulting in water discharge onto sensitive ICT equipment. Because of this wet pipe sprinkler systems are not suited for providing primary protection of ICT equipment, but are very useful to minimise and localise the impact of a fire. It should also be said the probability of water leaks occurring within sprinkler pipe work that has been properly installed and pressure tested is very low. Page 7 of 18

8 Fire Australia 2010 Practical Fire Protection in Computer Room and Data Centres In a data centre, if the data halls are not sprinkler protected then the remainder of the building should be sprinkler protected, to minimise the potential for a fire outside the data hall(s) impacting on the data hall. This has been evidenced in some notable fire events in recent history, such as 1975 NY fire, 1988 Hinsdale Central Office fire and the 1994 Pacific Bell Los Angeles telephone exchange fire. All of these fires resulted in significant interruptions to telephony services, including emergency services, which may have been reduced in impact had wet pipe sprinklers been installed. Due to these type of events, the importance of wet pipe sprinklers is recognised in data centres and major telephone exchanges, with most modern data centres including installing wet pipe sprinkler protection in all ancillary areas and sometimes within the data halls themselves. When it comes to Computer rooms they are often located within existing office area to provide ICT services to that site and/or subsidiary branch site. In many cases such as large warehouses or offices within high-rise buildings the existing facility is fitted with a wet pipe sprinkler system as a base building fire protection system. Where this occurs there is often a request to remove sprinkler protection from the computer room to avoid the potential for water damage. As mentioned previously the fears of water leakage are often misplaced - in a well installed and commissioned system, water leakage is a very rare occurrence. There are some practical measures that can be taken to retain wet pipe sprinklers within the computer room while minimising the potential for water leaks. ACCIDENTAL BREAKAGE OF SPRINKLER HEADS Where computer rooms are retrofitted within an existing site that has wet pipe sprinklers installed it is common to field a request to plug off the sprinkler heads to minimise potential for water damage due to accidental actuation of a sprinkler head. A practical solution to retain the benefit of sprinkler protection would be to install flush plate heads which are significantly less susceptible to accidental damage. If the computer room is not fitted with a ceiling, then upright heads and sprinkler head cages can be used to minimise accidental mechanical contact with the sprinkler element. USE OF HIGH TEMPERATURE HEADS Sometimes technical specifications call for existing sprinkler heads within computer rooms to be replaced with high temperature heads ( eg 79 C / 93 C rated heads). There is no documented philosophy behind this approach. Presumably it is thought that either higher temperature heads are more resistant to physical damage or that there is a risk from over temperature due to cooling system failure. In the first case, high temperature heads are equally susceptible to breakage as low temperature heads so if this is the reason then installing high temperature heads is providing false comfort. In the case of potential activation due to over temperature, given that standard detector heads don t operate until the sprinkler element reaches 68 C; and by this time the air Page 8 of 18

9 Practical Fire Protection in Computer Room and Data Centres Fire Australia 2010 temperature is much warmer; then there are likely to be risk to the equipment regardless of sprinkler activation or not. To address this risk, rather than installing high temperature heads, it would be better to invest in an environmental monitoring system which can notify on and off site staff should the room temperature exceed specified limits and may even be programmed to initiate shutdown of equipment if the temperature keeps rising. LEAKAGE FROM SPRINKLER PIPE WORK Besides water damage due to leakage from an accidently actuated sprinkler head, water may potentially leak from sprinkler pipe work. However in a properly installed and commissioned system the probability of leakage is relatively low. Nonetheless to minimise the potential there are a number of things that can be done, such as: Replacing the existing sprinkler pipe work over the computer room with new pipe work Installing Galvanised pipe work (even for wet pipe systems) Pressure testing the section of pipe work over the computer room (can be difficult in an existing installation). In addition to the above, for smaller computer rooms with 2 rows of heads or less, the mains feeding the sprinkler heads within the computer room may be relocated outside the room boundaries with single piece flexible sprinkler droppers from the mains to the head. This is shown schematically below in Figure 2 & Figure 3. Due to the lack of screwed joints and stainless steel construction the flexible dropper is unlikely to leak either. Should the main develop a leak either due to corrosion over time or at a joint then there is unlikely to be direct damage to equipment. Sprinkler main / range pipe One Piece flexible sprinkler droppers Below Ceiling Sprinkler heads Computer Room Figure 2. Possible configuration of sprinkler system to minimise potential for damage due to pipe leaks Page 9 of 18

10 Fire Australia 2010 Practical Fire Protection in Computer Room and Data Centres Sprinkler main / range pipe One Piece flexible sprinkler droppers Below Ceiling Sprinkler heads Computer Room Figure 3 Alternative arrangement using sidewall sprinkler heads INSTALLATION OF ACCESSIBLE ISOLATION VALVES Where wet pipe sprinkler are installed over critical ICT equipment or ancillary plant (eg UPS) it is recommended that an isolation valve be installed in an readily accusable location so that in the event of actual or accidental operation of a head the water can be shut off relatively quickly to minimise water damage. The valve should be secured to prevent un-authorised operation but staff assigned to respond to fire events within the facility should be trained and equipped with keys and/or other relevant access provisions. PROFESSIONAL REMEDIATION SERVICES In the very rare event that something goes wrong and water is discharged then all is not lost. There are a number of professional remediation services available that can restore data from equipment subject to water damage as well as other types of damage including soot from a fire. Thus it is highly recommended that such services be included as part of a disaster mitigation and recovery plan. PRE-ACTION SPRINKLER SYSTEMS Pre-action sprinkler systems are becoming more prevalent both in data centres and computer rooms. Often when a computer room is installed as part of a total office refurbishment a tail-end pre-action system is installed to protect the computer room, in lieu of wet pipe sprinklers. The increase in popularity of pre-action systems within data halls and computer rooms is largely attributed to the development of double interlock pre-action systems which require two forms of detection prior to water being admitted into the system pipe work. The most common configuration Page 10 of 18

11 Practical Fire Protection in Computer Room and Data Centres Fire Australia 2010 utilised in data centres is electric/pneumatic double interlock, which requires operation of both the fire detection system and loss of pneumatic pressure in the system pipework (due to sprinkler head operation). This means that a false alarm on the detection system by itself will not cause the pipework to flood, similarly if a sprinkler head is accidently broken water will not be released unless the detection system is activated also. While pre-action systems can provide the benefits of a wet pipe sprinkler system, with reduced potential for water leakage, there are a number of issues, that if not addressed, can erode this benefit and make the pre-action system problematic. DRAINAGE OF PIPE WORK Even though it is not intended that the pre-action system pipework have water admitted unless there is a fire, there is always a chance that at some point the system will be activated when there is not a fire and the pipe work will need to be drained. The Australian Standard AS requires that pre-action pipe work be installed on a gradient to drain back toward the stop valve, however there are instances where low points occur in pipework prior to the valve, such as where beam wraps occur. All low points should be fitted with a normally locked drain point. One other complication that arises is installation of pre-action systems in data centres and computer rooms fitted with false ceilings. This necessitates installation of droppers to serve each sprinkler head, thus should the system need to be drained each sprinkler head needs to be removed so that the water can be drained out potentially splashing water on equipment below as well as being a very time consuming process. Alternatively flexible sprinkler droppers may be used so that if they need to be drained they can be flexed up without removing the head, however this is even more time consuming than removing the head. Thus wherever possible all sprinkler heads on a pre-action system should be configured as upright heads to significantly simplify the mean time to re-instatement of the pre-action system should it be accidently be activated. PRESSURE TESTING OF PIPE WORK Often many pre-action systems are installed without the pipe work being pressure tested due to fear of water leaking on ICT equipment being installed at the same time. Furthermore pressure testing the pipework with water means time consuming draining and drying operations. While it is possible to pressure test with air, there are Occupational Health and Safety concerns with inflating pipe work with air to pressures of around 1700 kpa. And there is no clear guidance available on using lower air pressures to gain an equivalent test. This means that the first time that the pipework is pressure tested is if an accidental discharge occurs. Obviously it is far better to have this occur, before equipment is moved into the room, Page 11 of 18

12 Fire Australia 2010 Practical Fire Protection in Computer Room and Data Centres therefore it is highly recommended that time is allowed to carry out the pressure test or at the very least actuate the system. USE OF CORROSION RESISTANT PIPE WORK Australian Standard AS does not have any specific requirements in relation to pipework for pre-action systems other than the same requirements that apply for wet pipe systems. While it is possible to pressurise pre-action systems with inert gas such as Nitrogen, for simplicity the vast majority of pre-action systems are pressurised with ambient air via an air compressor. Thus if standard black steel pipe is used the moisture in the air will corrode the pipe internally. Thus all pipework should be corrosion resistant internally as well as externally for black steel pipe this typically means a hot dip galvanisation for each piece of pipe. Alternatively other types of pipe, such as stainless steel may be used. INSTALLATION OF STOP VALVE ABOVE PRE-ACTION VALVE Although not required by AS2118.1, it is strongly recommended that an isolation valve be installed downstream of the pre-action valve to allow regular (weekly/monthly) testing of the valve without introducing water into the system pipe work. GASEOUS SUPPRESSION SYSTEMS Gaseous suppression systems are typically installed in data halls or network operation control areas of data centres, however there is an increasing trend for installing gaseous suppression within computer rooms. Gaseous suppression systems are typically based on oxygen displacement by inert gas or chemical suppression of flame, however it is not the intent of this paper to discuss the advantages and disadvantages of either approach. Gaseous suppression systems are generally considered to be the optimal fire protection solutions for computer rooms and data centres as they can extinguish an incipient fire, preventing spread to adjoining equipment in the same rack, with minimum clean up. Thus apart from the equipment affected by the fire, the ITC facility may not even shut down or could be back up and running within minutes after discharge of the gas. Even though gaseous suppression systems can be extremely effective and are relatively costly there is a tendency to think that they are bulletproof and don t need much consideration other then the decision whether or not to install such as system. However there are many factors which can affect the efficacy of a gaseous suppression system. Page 12 of 18

13 Practical Fire Protection in Computer Room and Data Centres Fire Australia 2010 EXTENT OF PROTECTION As documented by Robin & Forssell [3], gaseous suppression systems are very effective at suppressing a fire within the area they protect, however provide effectively no protection for fires beyond their enclosure or area of operation. This is important to remember in the context of a computer room located within an office. Often businesses will commit to significant expenditure installing a gas suppression system within a computer room with office space that has no sprinklers or automatic fire detection. Often retrofitting a building wide system is well outside the budget in such cases where this cannot be achieved consideration should be given to improving the passive fire protection of the computer room s bounding construction. At a minimum, walls should be extended above the ceiling to minimise smoke ingress into the computer room for as long as possible. TOTAL FLOODING VS LOCAL APPLICATION Often gaseous suppression systems in ICT applications are thought of as being total flooding systems whereby the whole data hall or computer room is flooded with gaseous suppression agent to its extinguishing concentration. However this can be very expensive and in situations where the budget doesn t allow for total flooding consideration should be give to local application system(s) to protect just the critical equipment / racks within the data room, rather than the whole room. There are number of proprietary systems available for protecting individual ICT equipment. RESERVE AGENT Due to the high cost and storage space, reserve agent often gets omitted from new gaseous suppression installations. One of the benefits a gas suppression system provides is its minimal potential for collateral damage and cleanup providing the ability to get back up and running quickly. However after a discharge if provisions for reserve gas have not been made then it may take many days or even months to restore the gaseous suppression system back to operational status. Even if off-site reserve gas is available at short notice, the logistics of swapping out agent containers, particularly in large inert gas systems, can result in significant restoration times and thus significant period with the facility un-protected. Traditionally reserve agent is implemented as an active reserve whereby the reserve agent containers are plumbed into the discharge manifold and fitted with discharge actuator(s). In the event of discharge of the primary agent containers the reserve gas is simply brought online electronically via a switch at the control panel. While this approach enables rapid restoration of the gaseous suppression system there is extra cost in extending the discharge manifold and providing additional actuator(s). Page 13 of 18

14 Fire Australia 2010 Practical Fire Protection in Computer Room and Data Centres As an alternative, passive reserve gas can be provided whereby the reserve agent containers are stored with the primary containers but not connected. Should a discharge occur, the system can be restored relatively quickly by swapping out the discharged containers and connecting the reserve containers. Ideally this should be set up so the discharge tubing can be swapped from primary to reserve cylinders without manual handling to reduce changeover time. It should also be noted that recent federal legislation changes mean that if the agent is an ozone depleting substance or synthetic greenhouse gas that licensing of the system owner/operator may be applicable. FULL DISCHARGE TESTING There has been a recent trend away from full discharge testing, primarily due to cost constraints as well as the implication it has on construction programme. Nonetheless conducting a full discharge test is useful, not only to confirm that extinguishing agent concentrations are achieved, but to test the system and the room enclosure prior to fitout with ITC equipment. Similar to the discussion on full testing of pre-action systems, if a gaseous suppression system is not fully discharged during installation / commissioning, the first time it is discharged may be when the room is fully occupied with ICT equipment, potentially with catastrophic consequences. For example if pressure relief dampers have not been set correctly (or even installed) over pressure can occur causing the bounding construction to fail. Given the high cost of equipment, and cost of downtime if that equipment becomes non-operational, sometimes the seemingly high cost of discharge testing can become minor compared to finding out if there are problems if the system is discharge during operation. It is important to remember that under the governing Act [4] there are controls on discharging certain gaseous agents and a permit will be required to undertake a discharge test involving those substances listed in Schedule 1 (in relation to fire protection systems this primarily relates to systems utilising HFC s as the extinguishing agent). PASSIVE FIRE PROTECTION While much of the focus on data centres and computer rooms concerns active fire protection systems, the importance of passive fire protection measures should not be underplayed. Passive fire protection generally refers to smoke and/or fire resistance of the walls, floors and ceilings forming the bounding construction. Passive protection is important to minimise the effect of a fire outside the computer room / data hall having an impact on the ICT equipment within the data hall. Providing true fire and smoke separation of existing rooms can be problematic due to the extent of existing services such as ducts and pipes which are usually present in the vicinity of the room in Page 14 of 18

15 Practical Fire Protection in Computer Room and Data Centres Fire Australia 2010 question. Where full fire separation is not readily achievable consideration should be given to smoke separating and sealing the computer room from the remainder of the building as much as possible. Although smoke separation has limited fire resistance it can minimise the potential quantity of corrosive smoke which may enter the data room and cause damage to sensitive equipment. CABLE PENETRATIONS An important issue to consider in passive fire protection is cable penetrations, as data centres & computer rooms are all about processing of data which flows in and out of the facility via cables. These cables need to penetrate the bounding construction compromising its fire and/or smoke resistance. At the time of construction these penetrations are usually fire stopped to maintain the integrity of the wall, however over time this fire stopping often gets removed and/or new penetrations are made by cable installers who are often unaware that they are affecting a fire protection system. Thus strict control / management of cabling works needs to be undertaken during operation of the facility to ensure that the integrity is not compromised. Additionally proprietary cable firestop solutions are available which can simplify fire stop management. COMMISSIONING Proper commissioning of fire protection within ICT facilities is of paramount importance to ensure that the installed systems meet and continue to meet their design objectives. A well designed fire protection system can quickly turn into an ineffective system if not commissioned and maintained properly. The fire Industry is generally good at providing installers statements and generic commissioning checklists which state that everything complies, however the industry is usually poor at providing detailed commissioning records against which pro-active accurate inspection, testing & maintenance can be undertaken in the future. It is recommended that the following typical commissioning data be recorded (in addition to installer s statements and checklists required by the relevant installation codes & standards) as applicable: Point type fire detection systems - Type & Setting for each individual detector where applicable - Alarm threshold(s) for each detector - Operational matrix including multi-event outputs (eg dual zone operation) - Battery capacity calculations & verification - Verification record of system operation sequences. Aspirated Smoke Detection System - Transport time and/or static pressure at each sampling point. Page 15 of 18

16 Fire Australia 2010 Practical Fire Protection in Computer Room and Data Centres - Detector head alarm sensitivity threshold settings (incl day/nite settings) - Alarm threshold delays and type of delay (eg cumulative/simultaneous) - Battery calculations (if not part of above) Emergency Warning System - Type of each speaker - Power tapping of each speaker (as installed/commissioned) - Sound pressure levels for each type of alarm signal (eg alert / evacuation) at various locations - Size, configuration & loading of each amplifier Water Based Fire Protection Systems - System flow tests - Normal system static pressures Pre-action Sprinkler Systems - Supervisory air pressure settings - Compressor size (free air flow/hr complete with calculations Gaseous Suppression System - Room integrity test results, including estimated leakage area - Agent container liquid level / weights (recorded on the container) - Nozzle details for each discharge nozzle INSPECTION, TESTING & MAINTENANCE As with commissioning ongoing Inspection, testing and maintenance is critical to maintaining the performance of the installed systems, to ensure they have the best chance of effectively dealing with a fire when required. On-going inspection testing and maintenance should re-verify critical information required during commissioning so that early failure or system degradation can be detected. However, systems are often inspected and tested for compliance against the relevant or current installation standard, not what level of performance the installed system was commissioned to. For example Aspirated smoke detection systems installed to AS are required to have a transport delay of no more than 90s. If a system was installed and achieved a transport time of 30s at commissioning, but in subsequent years the transport time was 70s, it most probably wouldn t be recognised as a potential problem as the tester will simply be verifying against the minimum requirements of the installation standard. In a similar manner maintenance contracts for many small computer rooms do not include provision for annual room integrity testing, so it simply is not undertaken. As computer rooms are notorious for having uncontrolled cable penetrations being made during their lifetime, it is likely Page 16 of 18

17 Practical Fire Protection in Computer Room and Data Centres Fire Australia 2010 many of these penetrations will go unnoticed, significantly affecting the efficacy of the gaseous suppression system. Page 17 of 18

18 Fire Australia 2010 Practical Fire Protection in Computer Room and Data Centres REFERENCES [1] Mitchell, R.L., Data center density hits the wall, Computerworld, Jan , accessed at on 25/6/2010. [2] Mcubbin, A. The Australian Standard for Computer Accommodation, Business Data Strategy, October 2008, accessed at on 25/6/10 [3] Robin M.L., Forssell E.W., Comparison Testing in a simulated data processing / telecommunications facility: FM200 and automatic fire sprinklers, Halon Options Technical Working Conference, 14th. Proceedings. HOTWC [4] Ozone Protection and Synthetic Greenhouse Gas Management Act 1989 (CWealth) Page 18 of 18