What a capacity plan has to check together
IT load. Separate installed nameplate power, measured busy-hour load, committed additions and the growth assumption. Do not treat a nameplate as measured demand.
Racks and space. Count usable racks, power density per rack, floor loading, service clearance and aisle space. Spare floor area does not mean the target row has a place to install.
Power and redundancy. Check available capacity in normal state and after one path or one module is removed. Record A/B feeds, UPS, distribution and any upstream feeder that both paths share.
Network uplinks. Check the fabric and uplinks with busy-hour traffic and a growth forecast. The sum of port rates does not replace concurrent demand, and an aggregate does not guarantee the rate of one flow.
Cooling constraint. Hand the heat from IT and electrical loss to the cooling topic. A rated cooling tonnage does not by itself prove the rack inlet temperature.
How to name the bottleneck
- Define the boundary: room or row, racks, power unit, design interval and availability target.
- Write the current case and the growth case. Mark each number as measured, design or assumption.
- Calculate remaining space, power, network and cooling, including the failure state.
- Express each limit as the same planned load and record which system hits its limit first.
- Size the expansion for that bottleneck, then recheck the systems next to it.
Overall capacity is not the sum of ratings in different units. It is the largest load that still satisfies every constraint.
Worked example: power is short after growth
This is a planning example, not a site test or a procurement result. Twenty racks are planned at 8 kW each, so the current IT plan is 160 kW. Growth is 20 percent. Two independent power paths can each deliver 120 kW.
The growth case is 160 kW × 1.20 = 192 kW. If both paths share the load, each carries about 96 kW, which is under its 120 kW rating. That still leaves switching, balance and any shared upstream feeder to verify.
After one path is removed, 120 kW remains. The gap is 192 − 120 = 72 kW. If the rule is that one surviving path must carry the whole grown load, power redundancy is the bottleneck. Confirm that path before adding racks or servers. Cooling still needs its own check of the matching heat load and N+1 sensible capacity.
The example does not model power factor, UPS efficiency, derating, inrush, a feeder that both paths share, network traffic or the thermal environment. A real project should recheck with equipment curves and field measurements.
Continue with a specialist check
Network architecture inside the capacity plan
Infrastructure capacity includes racks, space, power and network dependencies. Choose an architecture from explicit requirements, team readiness and operating trade-offs, then check link capacity separately. A preference score does not certify hardware capacity or redundancy.
Engineering boundary
Use this page to screen a plan. It is not a certified design, a construction approval or a site acceptance. Distribution protection, structural loading, airflow, equipment derating and the redundancy topology have to be confirmed by a qualified engineer against the applicable standard and the site.
FAQ
How do I find the bottleneck in a data center capacity plan?
Check space, power, network and cooling against the same load and time boundary, including the failure case. The shared load all four can still carry is the plan limit.
Does spare cooling mean I can add servers?
Only if power, rack space and uplinks can carry the same addition. A surplus in one system does not move the limit set by another.