A Data Center Case Study: Why A/B Power Monitoring Can Give a False Sense of Security
Consider a data center where a rack has been allocated 10 kW of redundant IT capacity.
The rack is supplied through two independent power paths:
Power Path A → Rack PDU A
Power Path B → Rack PDU B
The design intention is straightforward:
If either power path fails, the surviving path must be capable of supporting the rack’s required critical IT load.
This is the essence of A/B power redundancy.
But a problem can appear when capacity monitoring focuses only on the individual utilization of PDU A and PDU B.
The Scenario
As servers and other IT equipment are installed, many devices are equipped with dual power supplies.
One power supply is connected to PDU A.
The second power supply is connected to PDU B.
During normal operation, properly designed dual-corded equipment may share its load between both power supplies.
The DCIM system may therefore display:
PDU A Load: 6 kW
PDU B Load: 7 kW
If each PDU is monitored independently against a 10 kW operational limit, both values appear acceptable.
PDU A:
6 kW / 10 kW = 60%
PDU B:
7 kW / 10 kW = 70%
Everything appears normal.
Both PDUs may even be shown in green on the monitoring dashboard.
But from a redundancy perspective, the rack is already overloaded.
The Real Rack Load
The critical IT load being supported by the rack is approximately:
Total Rack Critical Load = A Load + B Load
Therefore:
6 kW + 7 kW = 13 kW
The rack is carrying approximately 13 kW of IT load.
However, its allocated redundant single-path capacity is only:
10 kW
The important question is therefore not:
Is PDU A overloaded?
or:
Is PDU B overloaded?
The important question is:
Can either power path safely support the required rack load if the other path fails?
In this case, the answer is no.
What Happens When One Power Path Fails?
Assume Power Path A becomes unavailable.
Dual-corded IT equipment that is designed to continue operating from the remaining source will transfer or redistribute its power requirement to Power Path B.
Before failure:
A = 6 kW
B = 7 kW
After loss of A:
A = 0 kW
B ≈ 13 kW
The exact electrical value may change slightly because of PSU efficiency, load-sharing behavior and operating conditions, but the surviving path must effectively support the full required IT load.
The problem is clear:
Required Failover Load ≈ 13 kW
Allowable Single-Path Capacity = 10 kW
Therefore:
Single-Path Capacity Exceeded by approximately 3 kW
The same problem exists if B fails and the load transfers to A.
From Redundancy to Cascading Failure
If the surviving electrical path cannot safely support the transferred load, the failure sequence could become:
Power Path A fails
↓
Critical IT load transfers to B
↓
B becomes overloaded
↓
Protective device, branch circuit, rack PDU or upstream component may operate
↓
Remaining power path is lost
↓
Rack loses power
The physical design contained two power paths.
But operationally, the rack no longer had sufficient redundancy.
This is an important distinction:
Having two power feeds does not automatically mean the rack remains redundant.
The surviving path must have sufficient usable capacity to carry the required load.
The DCIM Was Not Necessarily Wrong
Suppose the DCIM displays:
PDU A: 6 kW — Normal
PDU B: 7 kW — Normal
The measurements themselves may be completely correct.
The problem is not necessarily the measuring device.
The problem may be the monitoring logic.
If the system evaluates only the utilization of each rack PDU independently, it can miss the real redundancy risk.
For an A/B powered rack, monitoring should consider both:
Normal operating utilization
and
Single-path failure utilization
A Better DCIM Calculation
For a rack designed to operate with redundant A/B power, the monitoring system should calculate:
Rack Critical Load = A Load + B Load
Using this example:
Rack Critical Load = 6 + 7
Rack Critical Load = 13 kW
Then compare the critical load against the usable capacity of each surviving power path.
If:
Allowable Single-Path Capacity = 10 kW
then:
Redundancy Headroom = Single-Path Capacity − Rack Critical Load
Redundancy Headroom = 10 − 13
Redundancy Headroom = −3 kW
The negative value immediately indicates that the rack can no longer safely maintain its intended A/B redundancy.
Failover Utilization
A useful DCIM KPI can be introduced:
Failover Utilization
Failover Utilization = Rack Critical Load ÷ Allowable Single-Path Capacity × 100
For this rack:
13 ÷ 10 × 100 = 130%
Therefore:
Failover Utilization = 130%
The operational status should not be shown as normal.
It should indicate something such as:
CRITICAL — SINGLE-PATH REDUNDANCY CAPACITY EXCEEDED
This gives operators a much clearer indication of the actual risk.
Example DCIM Display
Instead of displaying only:
PDU A: 6 kW — 60%
PDU B: 7 kW — 70%
the system should also display:
Allocated Redundant Rack Capacity: 10 kW
PDU A Load: 6 kW
PDU B Load: 7 kW
Rack Critical Load: 13 kW
Single-Path Capacity: 10 kW
Redundancy Headroom: −3 kW
Failover Utilization: 130%
Redundancy Status: CRITICAL
Now the operator can immediately see that although neither PDU is individually overloaded during normal operation, the rack cannot safely tolerate the loss of one power path.
Capacity Planning Must Follow the Same Principle
The same logic should be applied before installing new servers or other IT equipment.
The capacity-planning question should not only be:
Does PDU A have spare capacity?
or:
Does PDU B have spare capacity?
Instead, the question should be:
After installing this equipment, can the rack still support the required load following the loss of either power path?
For a rack with an allocated redundant capacity of 10 kW:
Total Critical IT Load should remain within the allowable single-path capacity.
Once the total critical load exceeds that limit, the rack has effectively exhausted its usable redundant capacity even if both rack PDUs still show available capacity during normal operation.
kW Is Not the Only Limit
There is another important engineering consideration.
Monitoring only total kW is not sufficient.
A real rack power-management system should also consider:
Current per phase
Phase imbalance
Rack PDU rating
Branch circuit rating
Breaker rating
Outlet limits
Upstream RPP/PDU capacity
UPS capacity
Power-path capacity under failure conditions
For example, a three-phase rack PDU may remain below its total kW rating while one phase is approaching its allowable current.
Therefore, redundancy calculations should be supported by electrical protection and phase-level monitoring.
The Same Principle Applies Upstream
The concept does not stop at the rack PDU.
The same failure-state capacity analysis should be applied throughout the power chain:
Rack PDU
↓
RPP / Floor PDU
↓
UPS
↓
Switchboard
↓
Transformer
↓
Generator
A redundant rack is not truly protected if the rack PDU can support the failover load but an upstream component cannot.
True redundancy therefore requires the entire surviving electrical path to have sufficient capacity following the expected failure condition.
Design for the Failure State
This case demonstrates an important principle in critical-facility engineering:
Redundancy must be evaluated under the failure condition, not only under normal operating conditions.
During normal operation, the infrastructure may appear healthy because the load is distributed across A and B.
But redundancy exists for the moment when one of those paths is no longer available.
The correct engineering question is therefore:
If either A or B disappears right now, can the surviving power path safely support the required critical load without exceeding electrical or operational limits?
If the answer is no, the intended redundancy has already been compromised.
The Bigger Lesson
A/B redundancy is not simply:
Two power feeds
It is:
Two independent power paths, each capable of supporting the required critical load under the defined failure condition.
Similarly, effective DCIM monitoring is not simply about collecting accurate electrical measurements.
It must interpret those measurements according to the design philosophy and redundancy architecture.
A dashboard can show:
A = Green
and
B = Green
while the rack’s actual redundancy status should be:
RED
That is why failure-state monitoring is essential.
Final Takeaway
Do not monitor only:
How much load is on A?
and:
How much load is on B?
Also monitor:
What load will the surviving path have to carry when redundancy is actually required?
For an A/B rack, consider monitoring:
A-side actual load
B-side actual load
Combined critical rack load
Single-path usable capacity
Redundancy headroom
Failover utilization
Phase and breaker loading
Upstream failover capacity
Because:
Normal operating conditions do not prove redundancy. The failure condition proves whether the redundancy is actually usable.
That is the difference between having redundant infrastructure on the drawing and having true operational resilience in the data center.
