A and B power feeds are supposed to be two independent paths from separate sources to the same equipment, so that a fault on one leaves the other carrying the load. The value lies entirely in the independence. Two cords from a dual-corded server that land on the same distribution unit, the same breaker, or the same upstream panel provide the appearance of redundancy and none of the substance.
Where the two paths are supposed to diverge
A properly designed arrangement separates the paths as far upstream as the facility can manage: separate distribution units in the rack, fed from separate branch circuits, from separate panels, from separate uninterruptible supply systems, ideally from separate transformers or utility services. Each additional shared element is a point where both paths fail together. Real facilities compromise somewhere, and that is acceptable as long as the compromise is known and documented rather than discovered during an outage. The critical discipline is knowing exactly where your two paths become one, because that point defines the largest single failure your equipment can survive. Facilities describe their level of separation in terms that sound absolute in marketing material and are rarely absolute in practice, which is why reading the actual single-line diagram matters more than the label.
How the paths quietly merge over time
Designs are clean and operations are messy. A rack fills up and a new server’s second cord goes to whichever outlet is free. A distribution unit is replaced and cords are reconnected from memory. A temporary extension is added during a migration and never removed. Someone rebalances load across circuits without knowing which circuit belonged to which path. Over a few years the as-built reality drifts from the drawing, and it drifts in the direction of less redundancy rather than more, because the easy mistake is always to use the nearest available outlet. Periodic physical verification is the only defense, and it means following cords by hand. Maintenance is the other driver for genuine separation. If the two paths are truly independent, one can be taken out of service for work while the other carries the load, which is worth as much as fault tolerance.
Each path must carry the entire load alone
If both feeds normally share the load, each is running at roughly half its capacity, and a failure of one instantly doubles the load on the other. That transfer must not overload the surviving path’s breaker. Facilities therefore limit normal loading on each path so that either can take everything, which means a dual-fed rack running each path near its limit is a rack that will trip a breaker the moment it needs redundancy most. This is a common and painful failure, because the trip happens during an event that already has attention focused elsewhere. Monitor per-path loading against the surviving-path case, not against the shared case. Per-path metering is what makes this visible, and configuring alerts against the surviving-path threshold rather than the breaker rating is the difference between a warning and a surprise.
Single-corded equipment in a dual-fed environment
Plenty of useful equipment has one cord. Feeding it directly from either path gives it no protection against that path failing. The usual answer is a transfer device that accepts both feeds and presents one output, switching between them when a source fails. These devices have their own considerations: transfer is not instantaneous, the connected equipment’s supply must ride through the gap, and the device itself is a shared component in both paths. Some equipment, particularly with active input current shaping, interacts poorly with a transfer event. Verify behavior by testing under realistic load rather than trusting the specification sheet alone. Where a transfer device is used, its own capacity and its failure behavior become part of the design, and it deserves the same testing regime as everything else rather than being treated as a passive accessory.
Verification that actually proves something
Documentation is a starting point, not evidence. Prove independence by tracing physically, labeling at both ends, and then testing: with the equipment’s owner aware and a maintenance window agreed, remove one path at a time and confirm nothing drops and that monitoring reports it correctly. Record the result with a date. Repeat after any rack change. All of this involves energized distribution equipment, so it is work for qualified personnel operating under the site’s electrical safety program, with appropriate protective equipment and procedures. Do not open panels or handle rack distribution wiring without that authorization and training. Label cords at both ends with the path they belong to, using a scheme that survives handling, because a label that falls off in a warm rack takes the verification effort with it.
What to check before you act
- Trace both cords physically and label them at both ends.
- Identify the exact point where the two paths first share equipment.
- Keep each path’s normal load below what one path can carry alone.
- Test one path at a time in an agreed maintenance window.
Common questions
Is plugging both cords into two outlets on the same strip redundant?
No. A fault in that strip or its upstream breaker takes both. The paths must diverge before they reach the rack.
How heavily can I load each path?
Only as heavily as the surviving path could carry on its own after a failure. Facility engineering sets the specific limit for your installation.
The practical takeaway
Find the point where your two paths merge, write it down, and keep each path loaded so either could take the whole rack. Then test it on a schedule, because redundancy that has never been exercised is an assumption wearing a diagram.
References and further reading
- NFPA 70E: workplace electrical safety practices
- OSHA Subpart S: electrical standards for general industry
- NFPA: understanding the National Electrical Code
Numerical scenarios are illustrative unless identified otherwise. Follow the exact product instructions; component ratings and local installation requirements can differ.
