Performance-based design and a fire and security strategy project

The earlier lessons of this course described the individual elements of data centre fire protection, life safety and security: codes and compartments, fire dynamics, detection, alarms, suppression, hydraulics, batteries, electrical hazards, physical and electronic security, and the maintenance that keeps them working. A real project must combine these elements into a single coherent strategy, and often has features that the prescriptive rules of the codes did not anticipate: halls larger than travel distance limits assume, contained aisles, rooftop battery systems, or suppression choices driven by business continuity rather than life safety. Performance-based design provides a disciplined way to show that such a strategy meets the codes' objectives. This lesson explains the performance-based approach and its main tools, design fire scenarios and the comparison of available and required safe egress time, then applies the whole course in a project to develop the fire, life safety and security strategy for a new data centre.

Prescriptive and performance-based design

Most buildings are designed to prescriptive requirements: the code states exactly what to provide (a maximum travel distance, a fire-resistance rating, a sprinkler density), and compliance is demonstrated by showing that each requirement is met. Prescriptive rules are simple to apply and enforce, and they embody long experience, but they cannot foresee every building. A performance-based design instead starts from the goals the code is meant to achieve, sets measurable criteria, and uses engineering analysis to show that a proposed design meets them.

The NBC and NFC are objective-based codes. Division B contains acceptable solutions, which are deemed to satisfy the code; Division A states the objectives (such as safety and fire protection of the building) and functional statements to which each acceptable solution is linked. A designer may propose an alternative solution, which must achieve at least the minimum level of performance required by Division B in the areas defined by the objectives and functional statements attributed to the acceptable solutions it replaces. The authority having jurisdiction decides whether to accept the alternative solution, usually on the basis of a report by a qualified engineer.

In data centres, performance-based approaches are commonly used to justify long travel distances in large halls, the omission or modification of sprinklers in specific rooms protected by clean agents, unusual battery installations, smoke control strategies that account for high cooling airflows, and fire separations that differ from those the occupancy classification would require.

The performance-based process

Guides such as the SFPE engineering guide set out a process with the following steps:

• Define the project scope: the building, its use, its occupants and the stakeholders (owner, insurer, AHJ, fire department, tenants).
• Identify goals: life safety of occupants and firefighters, protection of property, continuity of operations, and protection of the environment.
• Set objectives and performance criteria: measurable conditions that show each goal is met, such as tenable conditions along egress routes until everyone has left, no fire spread beyond the compartment of origin, or a maximum outage duration.
• Develop design fire scenarios and design fires (the heat release rate curves, smoke yields and locations to analyse).
• Develop trial designs and evaluate them against the criteria, using hand calculations, zone models, computational fluid dynamics and egress models as appropriate.
• Analyse uncertainty and sensitivity: how results change with different fire growth rates, occupant numbers or system failures.
• Document the design, its assumptions and the conditions on which it depends, and obtain approval.

The last step matters more than it may seem. A performance-based design depends on assumptions about how the building is used and maintained: occupant numbers, fuel loads, detection sensitivity, staff training. If those assumptions change, the design may no longer be safe. The documentation should therefore list the conditions that must be maintained, and the facility's change control process should check new proposals against them.

Tenability criteria commonly used for life safety include keeping the smoke layer above about 2 m, gas temperatures in the occupied zone below about 60 °C, visibility through smoke of at least 10 m in large spaces, and limits on exposure to toxic gases expressed as a fractional effective dose. The values chosen should be agreed with the AHJ.

Design fire scenarios

A design fire scenario describes a fire's location, the fuel involved, its growth rate and peak size, the state of the building (doors open or closed, cooling running or stopped) and the state of the protection systems. NFPA 101 Chapter 5 lists eight types of scenario that a design should consider, ranging from a typical fire in the occupied space to a fire in a concealed space, a slowly developing shielded fire, a fire near the largest fuel load, an outside exposure fire, and a fire with each active system unavailable in turn. For a data centre, a representative set might include:

• A fire in a server rack in a data hall with cooling running, detected by aspirating detection.
• The same fire with cooling stopped, which leads to a deeper smoke layer.
• Thermal runaway in a UPS battery rack, with off-gas release before flaming (Lesson 7).
• A fire in a switchgear room following an arc flash (Lesson 8).
• A cable fire in a concealed overhead tray or under the raised floor.
• A packaging fire in the loading dock or staging area, the largest fuel load in the building.
• A rack fire with the sprinkler system impaired for maintenance (Lesson 11).
• A fire in an adjacent building or yard (generators, transformers or vegetation).

Each scenario is analysed to see whether the performance criteria are met. Scenarios with low probability but high consequence, such as a fire with a protection system impaired, test the robustness of the strategy; the design need not perform perfectly in every case, but it should fail gracefully, for example by containing a fire within its compartment even if suppression fails.

Available and required safe egress time

The central life safety comparison in a performance-based design is between the available safe egress time (ASET), the time from ignition until conditions in a space become untenable, and the required safe egress time (RSET), the time occupants need to reach safety. The design is acceptable when ASET exceeds RSET by an adequate margin. RSET is made up of several phases:

where \(t_{det}\) is the time to detection, \(t_{alarm}\) the time from detection to the evacuation signal, \(t_{pre}\) the pre-movement time during which occupants recognize the alarm and decide to leave, and \(t_{move}\) the time to walk to an exit, including any queuing at doors. Pre-movement time is often the largest and most uncertain component; trained staff in a familiar building may respond within a minute or two, while visitors and contractors may take much longer. ASET may be estimated with a simple smoke filling calculation, which tracks the descent of the smoke layer as the fire plume pumps hot gases into it, or with a computer model.

import math

# ASET: time for the smoke layer to descend to 2 m in a data hall (cooling stopped, no venting)
# Plume mass flow (NFPA 92 axisymmetric plume): m = 0.071 Qc^(1/3) z^(5/3) + 0.0018 Qc
def aset(length, width, height, alpha, Qmax=1500.0, z_limit=2.0, rho=1.0):
    area = length * width
    z = height                      # height of the smoke layer interface above the floor
    t, dt = 0.0, 0.5
    while z > z_limit and t < 3600:
        Q = min(alpha * t * t, Qmax)        # kW, t-squared growth capped at Qmax
        Qc = 0.7 * Q                         # convective part
        m = 0.071 * Qc ** (1 / 3) * z ** (5 / 3) + 0.0018 * Qc if Qc > 0 else 0.0
        z -= m / (rho * area) * dt           # layer volume grows by m/rho
        t += dt
    return t

# RSET: detection + alarm + pre-movement + travel
def rset(t_detect, t_alarm, t_pre, distance, speed=1.0):
    return t_detect + t_alarm + t_pre + distance / speed

alpha = 0.01172                     # medium growth
for name, (L, W, H) in {"hall 40 x 30 x 4.5 m": (40, 30, 4.5), "hall 40 x 30 x 6 m": (40, 30, 6.0),
                        "battery room 6 x 5 x 3 m": (6, 5, 3.0)}.items():
    print(f"{name}: ASET {aset(L, W, H, alpha):.0f} s")
print()
t_asd = math.sqrt(10 / alpha)       # aspirating detection at about 10 kW
t_spot = math.sqrt(250 / alpha)     # spot detection in high airflow at about 250 kW (assumed)
for label, td in (("aspirating detection", t_asd), ("spot detection", t_spot)):
    r = rset(td, 60, 120, 70)
    a = aset(40, 30, 4.5, alpha)
    print(f"{label}: detect {td:.0f} s, RSET {r:.0f} s; 4.5 m hall margin {a - r:.0f} s (ASET/RSET = {a / r:.1f})")

hall 40 x 30 x 4.5 m: ASET 613 s
hall 40 x 30 x 6 m: ASET 744 s
battery room 6 x 5 x 3 m: ASET 54 s

aspirating detection: detect 29 s, RSET 279 s; 4.5 m hall margin 334 s (ASET/RSET = 2.2)
spot detection: detect 146 s, RSET 396 s; 4.5 m hall margin 217 s (ASET/RSET = 1.5)

With cooling stopped, the smoke layer in a 4.5 m high hall reaches 2 m after about ten minutes, and a higher ceiling adds two minutes more. With aspirating detection, occupants are out with more than five minutes to spare and ASET is more than twice RSET; with slower spot detection the margin falls by about two minutes. The battery room fills in less than a minute, which shows why battery rooms are designed to be unoccupied, with off-gas detection that warns before any fire and procedures that keep people out once an alarm occurs. The smoke filling model is conservative for a hall with cooling running, which dilutes the smoke, but the cooling may be shut down during an incident, and the analysis should cover both states.

The course project: a fire, life safety and security strategy

The project brings the whole course together. The brief is a new two-storey data centre with a total IT load of 24 MW in four data halls of 6 MW each, each about 3,000 m² with hot-aisle containment and a 4.5 m clear height. Each hall has its own electrical room with lithium-ion UPS batteries, and there are network rooms, a loading dock and staging area, offices for 40 staff, eight rooftop chillers, ten diesel generators in the yard with belly tanks, and two oil-filled transformers. The site has a municipal water main with a recent flow test of 5.5 bar static and 3.8 bar residual at 3,000 L/min. The owner wants to keep any single fire event to one data hall, with no data loss, and the insurer requires aspirating detection and pre-action sprinklers.

The strategy report should address each of the following, drawing on the corresponding lessons:

• Codes and compartments (Lesson 1): occupancy classification, construction, fire separations between halls, electrical rooms, battery rooms and support spaces, fire-stopping and egress layout.
• Fire risk assessment (Lesson 2): a register of fire scenarios with frequencies, consequences and the protection layers that address each.
• Detection (Lesson 3): aspirating system layout for halls with high airflow, alert levels, and detection in electrical and battery rooms.
• Alarms and emergency planning (Lesson 4): fire alarm architecture, notification inside containment, the sequence of operations matrix, and the outline of the fire safety plan.
• Suppression (Lesson 5): the choice between pre-action sprinklers, clean agents and water mist for each type of room, with clean agent quantities for the network rooms.
• Hydraulics and water supply (Lesson 6): a design area calculation, comparison with the flow test, and the need for a fire pump or tank.
• Batteries (Lesson 7): battery room design, off-gas detection and ventilation, explosion control and fire department information.
• Electrical fire risk and life safety power (Lesson 8): thermography and monitoring, arc flash reduction, emergency lighting and the separation of life safety power.
• Physical and electronic security (Lessons 9 and 10): security zones, access control and multi-factor points, video and storage, and the segmentation of building and security networks.
• Operations (Lesson 11): the ITM program, impairment procedure, hot work permits and records.
• Performance-based elements (this lesson): any alternative solutions, with ASET and RSET analysis.

Part of the strategy is choosing among protection options with different costs and benefits. A simple comparison annualizes each option's capital cost with a capital recovery factor and adds its maintenance cost and its expected annual fire loss.

# Comparing fire protection strategies for a data hall over a 20-year life
# option: capital cost ($), annual ITM cost ($), fire frequency (per year), loss per fire ($)
rate, years = 0.06, 20
crf = rate / (1 - (1 + rate) ** -years)          # capital recovery factor
options = {
    "A: pre-action sprinklers only":            (1.8e6, 40_000, 0.010, 9.0e6),
    "B: A + aspirating detection":              (2.4e6, 60_000, 0.010, 2.5e6),
    "C: B + clean agent in network rooms":      (3.3e6, 85_000, 0.010, 1.9e6),
    "D: B + battery room off-gas and separation": (2.9e6, 70_000, 0.007, 1.8e6),
}
print(f"capital recovery factor at {rate:.0%} over {years} years: {crf:.4f}")
best = None
for name, (capex, itm, freq, loss) in options.items():
    annual = capex * crf + itm
    risk = freq * loss
    total = annual + risk
    print(f"{name:<44} cost ${annual:>9,.0f}  risk ${risk:>8,.0f}  total ${total:>9,.0f}")
    if best is None or total < best[1]:
        best = (name, total)
print(f"lowest total annual cost: {best[0]}")
# Sensitivity: what if fires are three times as frequent as assumed?
for name, (capex, itm, freq, loss) in options.items():
    print(f"  x3 frequency, {name[:2]} total ${capex * crf + itm + 3 * freq * loss:>10,.0f}")

capital recovery factor at 6% over 20 years: 0.0872
A: pre-action sprinklers only                cost $  196,932  risk $  90,000  total $  286,932
B: A + aspirating detection                  cost $  269,243  risk $  25,000  total $  294,243
C: B + clean agent in network rooms          cost $  372,709  risk $  19,000  total $  391,709
D: B + battery room off-gas and separation   cost $  322,835  risk $  12,600  total $  335,435
lowest total annual cost: A: pre-action sprinklers only
  x3 frequency, A: total $   466,932
  x3 frequency, B: total $   344,243
  x3 frequency, C: total $   429,709
  x3 frequency, D: total $   360,635

On the base assumptions, the cheapest option (pre-action sprinklers alone) has the lowest total annual cost, narrowly ahead of adding aspirating detection. If fires are three times as frequent as assumed, aspirating detection becomes clearly worthwhile, and the battery room measures of option D also pay off. Since the fire frequencies are uncertain, the insurer requires aspirating detection anyway, and the comparison leaves out life safety, reputation and contractual penalties for outages, a sound recommendation is option D, with the clean agent rooms of option C considered only where specific equipment justifies them. The calculation does not make the decision, but it makes the trade-offs explicit and shows which assumptions matter most.

Professional practice

Fire protection and life safety design affects people's lives, and it is regulated accordingly. Designs must be sealed by licensed professional engineers where provincial or state law requires, and engineers must practise within their competence, using specialists for areas such as computational fluid dynamics or explosion protection where needed. Performance-based designs benefit from independent peer review. Early and continuing engagement with the AHJ, the fire department and the insurer avoids late surprises and builds the trust that alternative solutions depend on. Codes and standards change on regular cycles (the 2025 editions of the Canadian national codes, and three-year cycles for NFPA standards and the ICC codes), and the lessons of incidents such as the data centre fires discussed in Lesson 1 continue to shape them. Keeping up with these changes, and with new technologies such as high-density liquid cooling and new battery chemistries, is part of the profession.