Fire Protection In Data Centers
Fire protection for modern data centers is complex. The overall protection program needs to be informed by acceptable risk and meet the rigors of reliability and business continuity goals. Comprehensive protection programs developed to address expected fire risks provide a robust approach towards achieving these goals, rather than simply meeting local codes and regulations.
Protection goals and objectives for data centers include life safety, property protection, and business continuity. Life safety is mandated by codes and standards and includes providing safe exits, suppression, and adequate warning of fire or other hazardous conditions within the data center space. Property protection and business continuity goals are largely based upon owner, user, and other stakeholder requirements. These often include minimizing equipment damage and maintaining business continuity.
There are numerous detection considerations and approaches available for use within data centers to provide early warning of a fire. The inherent criticality and essential nature of the data center equipment will often dictate the detection approach. Detection strategies typically range from spot-type smoke detectors, air-aspirating smoke detectors (e.g., smoke sampling chamber with a sampling tube network), or a combination thereof. As with any approach, there are advantages and disadvantages to these strategies. The data center design and stakeholder goals help to guide the detection design.
Spot-type smoke detectors are a very common smoke detection strategy for data centers. They are often an inexpensive and simple solution. Cross-zoned smoke detection is typically the preferred strategy when utilizing spot-type detectors in data centers. This design relies upon the activation of two alarms before subsequent action such as opening of a pre-action valve or clean agent discharge. For addressable systems, a counter system can be used in which any two detectors will activate the sequence. Cross-zoning or counter system strategies can minimize the potential for an unwarranted discharge of a fire suppression system. The initial detector can provide a warning to operators and staff within the data center. The need for two separate detectors to activate, however, results in activation delay. This delay may compromise property protection and business continuity goals. Another concern is maintenance as the system presents a higher failure risk due to lack of maintenance.
Air-aspirating or air-sampling detection is becoming increasingly popular in data center applications. This type of detection is known for its ability to detect a fire in its incipient stages and therefore provides the faster response time and earlier warning than traditional detection. Similar to cross-zoned spot type detectors, air-aspirating detection has the ability to detect a fire at differing smoke obscuration thresholds and can therefore provide multiple warning levels. For example, a very low obscuration threshold can be programmed to activate a pre-alarm signal alerting staff to investigate and take action if needed. If a higher obscuration is subsequently detected, suppression systems can be set to discharge.
Detection above and below ceilings may also be a consideration, depending on the equipment and combustibles within those spaces. A qualitative risk analysis identifying the function of the particular data center will often guide the detection strategy utilized.
Design of fire protection systems in data centers needs to be carefully balanced with code requirements and stakeholder objectives. The primary fire protection systems used within data centers typically include: wet pipe sprinklers, pre-action sprinklers, and special suppression (e.g., clean agent, inert gas, or mist). Suppression systems need to consider higher challenge areas such as automated information storage systems (AISS) units and tape libraries.
Clean agent fire suppression systems are an option often used for property protection within a data center. These systems are classified as either halocarbon agents or inert gases. Both types of agents require a minimum design concentration based on the particular agent classification and potential fire scenario in order to extinguish a fire. Considerations such as pre-discharge warning signals, manual discharge stations, space considerations for agent supply, reserve agent supply, total flooding, or local flooding applications need to be considered when designing clean agent systems for data centers.
Fire suppression can be accomplished through either single-interlock or double-interlock pre-action systems. Single interlock systems rely on a separate event such as a smoke detector to activate before water is released into the system. If an individual sprinkler were to fail or break due to mechanical damage without a detector in alarm, the system will not release water into the pipes. After water fills the pipes, the system acts as a traditional wet pipe system and will not discharge water until temperatures in the room are high enough to activate a sprinkler. A double-interlock pre-action system provides an additional redundancy before water is released into the pipes. Both a detector actuation, typically a smoke detector, and a sprinkler actuation must occur simultaneously before water will enter the pipes. Critical applications can warrant a double-interlock pre-action system.
A wet pipe sprinkler system is the most basic sprinkler system option. This system, however, is always water filled, which is usually not preferred in a data center. These pipes have the potential for false discharge, leak, or breakage that could damage equipment and disrupt service. Welded pipe systems are preferable, although mechanical pipe joints are still found in many older facilities.
The Difference a Tier Makes
The Uptime Institute has developed an industry standard classification system. The simplest is a Tier 1 data center, which essentially includes non-redundant capacity and components. A Tier 4 classification is considered the most robust and less prone to failures. A Tier 4 design reduces downtime by employing fully redundant subsystems that are simultaneously active and compartmentalized. Once the tier level and associated tolerable risk and redundancy is understood, the team can tailor fire protection solutions to meet the stakeholder objectives.
The decision to use early and very early detection is typically based upon tier level, the criticality of the facility, and the risk tolerance of the stakeholders. For a business’ most critical operations, enhanced detection, either through very early detection or additional detection, can be important. For these cases, aspirated systems or multi-criteria detectors are a viable solution.
There are a wide variety of approaches to addressing issues of capacity, resiliency, cost, and schedule in data centers. Careful design that accounts for the range of possible fire scenarios allows the fire detection and protection systems to meet the reliability and business continuity goals and objectives. Commissioning is important, as designs can work well on paper but need to be properly implemented to meet the protection goals.
There are always a number of challenges with integration and coordination of components to fire suppression systems: the mechanical systems may need to be isolated and shut down, operators may need to receive pre-alerts, or the electrical equipment may need to be shut down. Each project is different and each facility includes different equipment and configurations. Data centers require comprehensive coordination. This is not necessarily limited to sprinkler and fire alarm, but also includes electrical, mechanical, and architectural components.
Communication is important. The design team needs to discuss stakeholder priorities to assess the acceptable level of risk in terms of property protection and business continuity. The team can then match the overall fire protection program, including features such as detection, suppression, protection of the room, and equipment resilience, with the stakeholder requirements. The design needs to be strategic; more may not necessarily be better. For example, additional detection beyond what is needed to meet stakeholder objectives could lead to unwanted alarms and associated downtime. The fire protection program needs to address how tolerant the stakeholders will be to unwanted alarms and the downtime associated with false alarms.
Designing to typical code requirements will not necessarily meet property protection and business continuity goals. Holistic designs that consider fire ignition scenarios and design for these scenarios, rather than simply meeting the code requirements, address these goals. Such holistic designs integrate systems into comprehensive fire protection programs, and incorporate requirements for building construction, fire- and smoke-rated walls and ceilings to protect the data center, interior finish limitations, egress, fire detection, and suppression systems. NFPA 75, Standard for the Fire Protection of Information Technology Equipment, outlines a comprehensive risk-informed approach to data center fire safety and protection.
Many data centers employ a pressurized raised floor to provide cold air to IT equipment, and an above ceiling area as a hot air return. Codes may require detectors under the floor or above the ceiling where HVAC piping, electrical feeders, or IT cables are placed within these plenum spaces. Due to the potential of fires originating within these spaces, detection is usually warranted within these areas.
Cold aisle thermal containment systems are also a “new” technology often employed within data centers to minimize the potential cooling loads within the space. These containment systems create a separate enclosure within the data center which must also be protected. In addition, the remaining hot aisle also can pose additional fire protection and fire alarm challenges. The temperatures within these hot aisles and immediately adjacent to the equipment can become hot. Temperature ratings of sprinklers and detectors needs to consider the expected ambient temperatures within the hot aisles. n