K&K Data Service Inc.Data Center Engineering Lab

K&K Data Service Inc.

Data Center Engineering Lab

Working analysis records from K&K's data center engineering program: failure domains, load arithmetic, cooling and hydraulic design, and the document discipline that holds them together.

This series publishes working engineering analysis from K&K's data center program. Numbers are labeled [TARGET], [MODELED], [SENSITIVITY] or [FACT]; nothing here is field-measured data, and no article is a sealed engineering document. Each article states its own limits and open items.

104 articlesSeries updated 2026-09-16
Aerial concept rendering of a single-story data center building on a wooded site at sunrise
Concept rendering from the K&K program’s customer materials.

Electrical Architecture and Failure Domains

  1. KK-DCL-001Two Feeders, One Substation: That Is Not Utility 2NBoth 12.47 kV feeders originate at the same substation, so the shared substation event deletes the nominal utility redundancy the one-line implies.
  2. KK-DCL-002Even Facility-Load Split Exceeds the 7 MW Feeder AllocationAn even normal-load split puts the 7 MW feeder over its limit by about 0.196 MW, so mechanical and house load must be biased to the 8 MW side.
  3. KK-DCL-003Single-Feeder Loss Is a Generation Event, Not a Utility TransferNeither surviving feeder carries the whole plant, so single-feeder loss is not a transfer event but an entry condition for controlled generation or load shed.
  4. KK-DCL-004UPS Sizing Follows Protected Load, Not the 12 MW Sales NumberUPS sizing follows the 12.711 MW protected-load figure that includes CDU and end-of-path losses, not the 12 MW billable IT headline.
  5. KK-DCL-005A Single 480 V Campus Bus Would Be 15,443 ACollapsing the campus onto one 480 V bus produces a current magnitude that the bracing, SCCR and arc-flash chain cannot absorb at concept stage.
  6. KK-DCL-006Each Dual-Cord Path Must Be Sized for Full Rack LoadEach cord of a dual-cord rack must carry full load under single-path operation, so conductor sizing does not halve when the second path is added.
  7. KK-DCL-007The 15 MW Service Cap and the 125% Conductor Rule Are Different LimitsThe 15 MW service cap and the continuous-load derating requirement are separate constraints that are frequently collapsed into one number.
  8. KK-DCL-008Matching MV Bus Identity to LV A/B Deletes Both Redundant PathsMedium-voltage bus ownership must be decoupled from the low-voltage A/B redundancy path or a single MV event erases nominal redundancy.
  9. KK-DCL-009Do Not Map Electrical Systems 1:1 onto Cooling ZonesAligning electrical failure domains one-for-one with cooling zones creates a coupled domain that fails both systems on a single event.
  10. KK-DCL-010Billable IT, Protected Load, Facility Load, and Service Cap DifferBillable IT, protected load, facility load and service cap are four different denominators that must be drawn on the same sheet to stop silent substitution.
  11. KK-DCL-011A 4 MW Zone Block Is 4,000 Billable, 4,237 Protected, 4,797 FacilityThe block named for 4 MW carries three different power figures — 4,000 kW billable, 4,237 kW protected and 4,797 kW facility — and only the first is a round number.
  12. KK-DCL-0128 MW and 7 MW Feeder Losses Leave Different Generation ShortfallsPath A and path B carry different mechanical and house allocations, so the two single-path failure cases are not mirror images.

UPS, Battery and Generation

  1. KK-DCL-013A Nine-Unit 3-to-Make-2 UPS Does Not Cover 12.711 MW Protected LoadA nine-unit 3-to-make-2 arrangement cannot deliver 12.711 MW of protected load, and the arithmetic failure is independent of vendor choice.
  2. KK-DCL-014Strict 2N UPS Misses 0.711 MW Protected Load in Maintenance-Plus-FaultStrict 2N leaves 12 MW against 12.711 MW when one path is in maintenance and the healthy path loses a single UPS module.
  3. KK-DCL-0153-to-Make-2 UPS Retains 14 MW in the Maintenance-Plus-Fault CaseThe twelve-unit distributed-redundant arrangement retains 14 MW in the same stacked maintenance-plus-fault case that defeats strict 2N.
  4. KK-DCL-016Catcher UPS Fails Maintenance, Transfer Refusal, and Dual-Block FaultThe catcher topology fails catcher maintenance, transfer refusal and simultaneous dual-block fault by construction, which makes it a cost reference only.
  5. KK-DCL-0173.3 MWh Battery Is 5.2 Minutes on 2N, Not 10 Minutes on Both PathsBattery energy divided across a 2N pair yields one autonomy figure, not two, and the doubling error survives into the battery room area.
  6. KK-DCL-018A 10× Recharge Rate Overruns the Last 4% of the 15 MW Service CapAggressive recharge current consumes the residual margin between facility load and the service cap, turning recharge into a capacity event.
  7. KK-DCL-019Battery Recharge Margin Cannot Be Counted Twice Against the Service CapThe 642 kW margin already contains charging power, so subtracting 233 kW again fabricates a 409 kW spare that does not exist.
  8. KK-DCL-020Low-Voltage UPS Topology Does Not Reduce On-Site Generator CountGenerator count follows facility load and site N+1, so changing the low-voltage redundancy topology does not reduce the generator quantity.
  9. KK-DCL-021Restoration Must Isolate Before Assuming Utility-Generator ParallelThe restoration sequence must isolate before it assumes a parallel utility-generator transition, because the parallel permit is not yet evidenced.
  10. KK-DCL-022A 0.5-Point UPS Efficiency Change Moves About 65 kW of Facility LoadA half-point change in UPS efficiency moves about 65 kW of facility power, against a non-critical utility line of 85 kW in the same schedule; the two are the same order, not equal.

Direct Liquid Cooling and CDU

  1. KK-DCL-023N+1 on the CDU List Is Not N+1 in the PipesAn N+1 equipment count does not survive contact with in-row zoning, valve granularity and pipe routing, which is where redundancy is actually decided.
  2. KK-DCL-024Eleven 1.5 MW CDUs Equaling 12 MW Is Coincidence, Not MarginEleven nameplate units happen to sum to the target, but the coincidence is not margin once redundancy groups and zone loads are applied.
  3. KK-DCL-025The Liquid Fraction f Mixes Design, Sensitivity, and OEM ValuesThe symbol f carries three non-substitutable calibres in one document - a hydraulic representative value, a set of sensitivity steps and an unobtained equipment-supplier input - and a neighbouring schedule uses two of them in adjacent columns.
  4. KK-DCL-026Why 100% Liquid Cooling Is Not a Design CaseThe 100 % row of the sensitivity table prints zero air-side heat and is labelled a theoretical limit; the same section bounds design at f no greater than 0.97 on an assumed 3 % to 8 % residual, and the white-space geometry is drawn to that bound rather than to the row.
  5. KK-DCL-027The 385 kW CDU Pump Budget Exists Only at a Rejected 6 K ΔTThe pump power budget that makes the PUE case work belongs to a delta-T that the hydraulic basis already rejected.
  6. KK-DCL-028PG25 Is Not Water Plus 4.5% VolumePG25 changes density, viscosity, specific heat and freeze behaviour together, so treating it as water with an additive is a property error, not a rounding one.
  7. KK-DCL-029A Water-Rated 1.5 MW CDU Is 1.20–1.28 MW on PG25Water-basis nameplate capacity derates on PG25, and the derate band is wide enough to change the unit count.
  8. KK-DCL-030Same-Day Plate Heat Exchanger Datasheet and Drawing DisagreeTwo vendor documents dated the same day carry the same model name under different item identifiers and different duty points.
  9. KK-DCL-03196 Isolation Valves Turn a Weld Repair Into Zero IT ImpactIsolation granularity, not spare capacity, decides whether a weld repair is a maintenance task or an IT outage.
  10. KK-DCL-032Automatic Leak Isolation Can Drop a 130 kW Rack on a False TripAn automatic isolation response to a leak signal without qualification can remove a loaded rack on a single false detection.

Hydraulics and Transients

  1. KK-DCL-033Raising Secondary ΔT From 6 K to 12 K Drops Two Pipe SizesDoubling secondary delta-T halves flow and moves the velocity screen by two nominal pipe sizes, which changes the structural and space case too.
  2. KK-DCL-034Isolation Raises Ring ΔP From 3.5 to 49.2 kPaThe isolation detour case, not the normal case, sets the ring pressure drop and therefore the pump duty point.
  3. KK-DCL-035Part-Load DLC Startup Runs Hotter Than Full Load If Valves ThrottleAt low load the flow regime leaves the turbulent band and the local film temperature can exceed the full-load case.
  4. KK-DCL-036Running Standby CDUs Saves 46% of Pump Energy, Not 56%Recommendation R-06's own factors — eight duty units' flow shared across eleven, each unit falling to 0.396 of full pump power — multiply to 0.545 of the duty-only baseline and a 46 % saving; the 0.436 and the 56 % printed beside them require a ten-unit baseline that the branch does not have.
  5. KK-DCL-037UPS Battery Minutes Do Not Equal Thermal Ride-Through SecondsElectrical autonomy and thermal autonomy are measured on different clocks, and the thermal clock is the short one.
  6. KK-DCL-038Electrical Ride-Through and Thermal Ride-Through Use Different ClocksPlotting ride-through seconds against coolant temperature rise on one figure shows where the two systems stop protecting each other.
  7. KK-DCL-039Steady Hydraulic Models Miss DLC Interlock, Inertia, and MaintainabilityThree unclosed items — leak interlock authority, thermal inertia and the parallel-maintainability service level — sit in the control, transient and contract domains, so the steady heat-balance and steady-hydraulic method that produced the rest of the basis cannot surface any of them.
  8. KK-DCL-040Put the Expansion Tank on Return, or Static Exceeds 600 kPaExpansion tank placement decides the static pressure reference, and the wrong side pushes local static above the component limit.

Heat Rejection and PUE

  1. KK-DCL-041Heat-Rejection Weather File Was Built for the Wrong StateA substation name was read as a city, the weather chain was built in the wrong state, and every downstream temperature became a template.
  2. KK-DCL-042786 kW Between PUE Metering Points Moves PUE by 6.6 PointsElectrical loss aggregated across transformers, UPS and distribution converts directly into PUE points once the denominator is fixed.
  3. KK-DCL-043Heat Rejection Is Not Secondary Loop HeatPrimary-side rejection adds air-side residual heat, primary pump work and an unquantified power-room share on top of the secondary loop figure, so sizing rejection plant on the secondary figure understates the duty by about 6.5 per cent.
  4. KK-DCL-044Secondary Supply Temperature Is the Entire Rejection ChoiceThe secondary supply temperature band spans nearly the whole feasible space of primary rejection routes, so freezing it freezes the route.
  5. KK-DCL-045Adiabatic N+1 Without Spray-Train Redundancy Reverts to Dry CoolingAn adiabatic unit counted as N+1 reverts to dry performance when the spray train has no redundancy of its own.
  6. KK-DCL-046100.1 m3/h at 348 kPa Is a Requirement, Not a Pump CurveThe review recorded a calculated duty point for each loop and no manufacturer curve, power, efficiency, suction-head requirement or nameplate, so no pump has been shown to meet it.
  7. KK-DCL-04750 kW of Fans Was Counted Twice in the Generator StepThe same 50 kW of air-side fan load was carried in two load-step registers, inflating the generator staging total and overstating the cooling side's share of it.
  8. KK-DCL-048A Modeled PUE of 1.177 Is Not Robust With a 275 kW ResidualA modeled PUE result carrying a residual term of a few hundred kilowatts is not a robust figure at the precision it is quoted to.
  9. KK-DCL-049Weather Margins Must Be Line Items, Not Hidden in Dry-Bulb or Wet-BulbDesign margin folded into dry-bulb and wet-bulb inputs cannot be audited, so margin must appear as its own line item.
  10. KK-DCL-050Do Not Freeze Heat-Rejection Model Numbers Before Supply Temperature ExistsReserving manufacturing capacity is reversible; freezing a model number before the supply temperature is decided is not.

Modules and Space

  1. KK-DCL-051White-Space Area Follows Geometry, Not kW per RackWhite-space area is driven by row geometry, aisle rules and service envelopes, so it does not scale linearly with rack power.
  2. KK-DCL-052Only 100 kW/Rack Divides 12 MW Into Three Equal ZonesAmong the studied density steps, only one divides the target load into three arithmetically equal zones without a remainder rack.
  3. KK-DCL-053Electrical Support Is 2.1–2.6× the White SpaceElectrical support rooms, clearances and maintenance aisles together occupy a multiple of the white space, not a fraction of it.
  4. KK-DCL-054Battery-Room Area Cannot Be Sized While Installed kWh Spans FourfoldWhile the battery energy requirement spans a factor of four, any single battery-room area is a selection, not a result.
  5. KK-DCL-055Three UPS Rooms Follow Failure-Domain Topology, Not Floor PlanThree UPS rooms follow from the failure-domain separation requirement, not from a floor-plan preference.
  6. KK-DCL-056Close the CDU and Rack Envelope Before Freezing Rack DensityRack and CDU physical envelopes bound the achievable density, so freezing density first inverts the dependency.
  7. KK-DCL-057Breaker Extract Depth Plus 0.9 m, Not Code Minimum, Sets the AisleThe chapter takes the front working clearance as the larger of the code minimum and the maximum extraction depth plus a 0.9 m transport margin, which is why the baseline low-voltage aisle is 1.50 m where the code allows 1.22 m.
  8. KK-DCL-058A BIM Object With a Port May Not Be Drawn as a 2D BlockAny object holding at least one declared port must exist in the model as a connectable instance carrying connectors; drawn as a block, a mass or a filled region it drops out of collision, quantity and topology checking without raising an error.

BIM and Navisworks

  1. KK-DCL-059Placing BIM Equipment Families Does Not Create Electrical SystemsPlacing equipment families creates geometry, not electrical systems; connectivity is a separate native object that must exist to be read back.
  2. KK-DCL-060BIM Electrical Circuits Can Report Zero Strictly Connected EdgesEvery one of the 78 direct connector-to-connector attempts was refused by the same native interface exception, and the circuit-semantics workaround that replaced it reads back 78 system objects and zero edges with both ends resolved on one system.
  3. KK-DCL-061BIM Semantic Readback Can Pass While the Evidence Chain Stays PartialSemantic readback can pass while the evidence chain remains partial, because readback tests the model and not the provenance.
  4. KK-DCL-062When Automated BIM Model Checks Pass but Human Review FailsA machine-validated drawing set passed every automated assertion and failed human visual review, which is why the visual gate is not optional.
  5. KK-DCL-063Thirteen Hard Clashes Were Real; the Zero-Result Test Is Not ClearanceThirteen hard clashes were genuine, and the third test's zero result proves only that the current proxy envelopes do not intersect, not that maintenance clearance has been approved.
  6. KK-DCL-0641,539 BIM Ports Are Endpoints, Not Connection Type CodesA large port count describes endpoints in the register and says nothing about how many distinct connection types are actually defined.
  7. KK-DCL-065BIM Electrical Systems Can Exist With Zero Connector Cross-ReferencesSystems can be created in quantity while the connector cross-reference set stays empty, which means nothing is natively joined.
  8. KK-DCL-066One Equipment ID Must Survive Model, BOM, and CommissioningA single equipment identifier must survive model, bill of materials and commissioning card, or traceability is reconstructed by hand every time.
  9. KK-DCL-067A BIM Purity Gate Stopped Federation at 80 of 96 Electrical InstancesThe purity gate stopped the next federation revision before any clash test ran, because the electrical model exposed 80 of 96 controlled instances; a clash result could never have reported connectivity anyway.
  10. KK-DCL-068Every BIM Output Needs Evidence, Route, and Procurement LabelsEvery BIM output carries a maturity label, because an unlabeled model view is read by downstream parties as a decision.

Controls, EPMS, BMS and DCIM

  1. KK-DCL-069Five Control Boundary Defects Trigger Together on Utility LossFive of the six named control boundary defects are triggered by the same utility-loss event, so they cannot be scheduled as independent fixes.
  2. KK-DCL-070Battery Charge Inhibit Must Be Hard-Wired, Not Supervisory-OnlyThe charge inhibit must exist in hard-wired logic, because a supervisory-only interlock is lost in exactly the scenario it protects.
  3. KK-DCL-071Black Start Needs Battery Charge While the Interlock Forbids ItBlack start requires charging power at the same moment the protective interlock forbids charging, and the conflict has no default resolution.
  4. KK-DCL-072A BMS Point List Without Quality and Command Feedback Is Not ReviewableWithout quality flags and command feedback fields, a point list cannot be reviewed for failure behaviour at all.
  5. KK-DCL-073Plant Liquid-Cooling Sequences Are a Separate BidPlant-level liquid cooling sequences cross three vendor scopes, so they belong in their own package rather than inside any equipment supply.
  6. KK-DCL-074Restoration Sequences Must Advance on Criteria, Not Fixed DelayEach state transition needs a measurable criterion and a timeout, because a fixed delay stamps success onto an unverified condition.
  7. KK-DCL-075A Common Time Base Does Not Recover Order in Joint-Test Alarm FilesA common time base is necessary for joint-event replay and not sufficient for it, because the delivered alarm matrix stamps annunciation and its own delays span zero to thirty seconds.
  8. KK-DCL-076Redundant Pairing, Not Platform Split, Doubles Monitoring Panel CountThe space model attaches the doubling of the monitoring panel count to redundant pairing in both of its statements and never to splitting the platforms.

Fire, Structure, Site and Compliance

  1. KK-DCL-077Fire Code Matrix Grades 20 Questions Empty; the Summary Says 18Twenty of the twenty-six code and authority questions carry no obtained input in the matrix itself while the chapter's own summary reports eighteen, and the register still separates which of them block work from which are merely unanswered.
  2. KK-DCL-078A PG25 Release Is Graded an Electrical Fault Before a FireThe fire basis files a glycol release as a release scenario and grades its electrical consequence above a combustion severity that cannot yet be assessed.
  3. KK-DCL-079Electrical Trip, Valve Close, and Zone Stop Are Separate Interlock EffectsElectrical trip, valve closure and zone shutdown are three separate effects that a cause-and-effect matrix must not conflate.
  4. KK-DCL-080Nonstructural Anchorage, Not the Frame, Is the Seismic RiskThe governing seismic risk sits in nonstructural component anchorage, which is specified by others and inspected last.
  5. KK-DCL-081A Uniform Floor-Load Target Does Not Bound Equipment Point LoadsA uniformly distributed floor target does not bound the point loads under equipment feet, and the ratio between them is large.
  6. KK-DCL-082UL Components Do Not Make a Listed AssemblyComponent listings do not transfer to the assembled enclosure, and the distinction decides who carries the compliance risk.
  7. KK-DCL-083Two Feeders in One Fire Area Is a Second Common CauseRouting both feeders through one fire compartment adds a second common-cause mechanism on top of the shared substation.
  8. KK-DCL-084Fuel and Coolant Bunds Cannot Be Counted as Stormwater DetentionSecondary containment volume cannot be counted as stormwater detention, and correcting the double-count tightens the land trial.

RFQ, Vendor Review and Commissioning

  1. KK-DCL-085The CDU RFQ Freezes a Four-Tuple Rated Capacity It Never DefinesThe procurement package names a four-tuple rated capacity as a defining parameter twice and never says which four quantities it holds, while its own card for that equipment lists six.
  2. KK-DCL-086Rejected Nine-Unit UPS Is Excluded in Quantity Rows, Not Binding Bid TermsThe rejected nine-unit arrangement is excluded in two rows that state a quantity and not in the row a bidder treats as binding, so the exclusion travels only as far as those two rows are copied.
  3. KK-DCL-087FAT Items Marked Site-Invisible Cannot Wait for SATAny characteristic that cannot be observed after installation must be witnessed at the factory, not deferred to site acceptance.
  4. KK-DCL-088A Winter Power-Failure Test Is Not Summer Rejection EvidenceA power-failure test executed in winter conditions does not evidence summer heat-rejection capability, whatever the pass stamp says.
  5. KK-DCL-089Integrated System Test Scenarios Are Named Before the Plant Is FrozenNaming integrated system test scenarios early forces the design to state which failures it claims to survive, before equipment selection closes it.
  6. KK-DCL-090Fourteen Transformers Are in the Test Set; the Count Is Not FrozenThe delivered commissioning test set sizes its object scope with the transformer count of one unfrozen topology branch of three, and its own first case is the version-lock that branch has not had.
  7. KK-DCL-091Missing Criteria Needs Its Own Root-Cause Class, Not the Commissioning OneThe defect index gives criteria-missing a root-cause class of its own, separate from both design and commissioning, because its corrective action is to obtain the criterion rather than to repeat the test.
  8. KK-DCL-092Joint IST Needs Electrical and Cooling Release StampsA joint integrated system test cannot be released by one discipline, because the failure it tests belongs to neither alone.

Errata and Open Questions

  1. KK-DCL-093Heat-Rejection Site Was Taken From a Substation Name, Then ReversedThe discrepancy between a substation name and the project's own state was raised as a change candidate at the first revision, and the ruling required the chapter holding the correct state to alter its wording.
  2. KK-DCL-094Two Records Ban Double-Counting 642 kW Recharge, Then Disagree What It IsThe shared project memory and the errata chapter both forbid subtracting the charging term twice, and read the figure it was subtracted from as headroom and as a branch load, so the correction that follows points opposite ways.
  3. KK-DCL-095A Model Audit That Verified the Wrong RepresentationVerification effort was spent with high precision on a representation that was never the engineering object under question.
  4. KK-DCL-096A CDU Branch Label Said 11 Units; the Instance Count Was 12A branch label and the underlying instance count disagreed, and the label was the one that travelled into downstream documents.
  5. KK-DCL-097User-Confirmed 155 kW Cabinet Load Has No Owner Approval RecordA vendor document carried a user-confirmed load figure for which no owner approval record exists on our side.
  6. KK-DCL-098Raised CDU Flow Was Square-Law Extrapolated From One Published PointThe operating flow was raised 26.742 per cent and the matching branch resistance was produced by a square-law extrapolation from the single published flow point, in a package whose only supplied curve carries no equipment identity.
  7. KK-DCL-099Four Engineer-of-Record Requests Have No Prerequisite; Three Need the OwnerOf twenty-one drafted engineer-of-record requests, exactly four depend on no other external request, and three of those four name the owner rather than an engineering discipline as the approving party.
  8. KK-DCL-100The 10-Transformer 4+1/4+1 Claim Was Struck, Not the Unit CountA ten-transformer arrangement was struck off the architecture gate's list of already-excluded expressions, on the ground of the availability claim attached to it, while the same unit count survives a few rows away as the cost comparator.

Hydraulic Calculation Detail

  1. KK-DCL-101Correcting CDU Pump Heat Moves the Close Threshold, Not the FailureSubstituting the chapter's own later secondary pump figure into its own duty identity moves the closing threshold from about 0.97 to about 0.99 and the published whole-site flow from 1,069 to about 1,044 cubic metres an hour, and the branch still fails to close.
  2. KK-DCL-102Three Hydraulic Flow Paths Closed; Residual Risk Left the ArithmeticA dimensional cross-check named the three unit confusions it existed to exclude, recomputed the flow basis by three paths that share no intermediate, and by closing them moved the chapter's residual uncertainty entirely onto four registered blockers.
  3. KK-DCL-103One Cabinet Min-Flow Percentage Describes Two Reynolds RegimesThe single percentage written into the control sequence as the cabinet branch minimum puts the branch at about 21,000 Reynolds at the design mean and about 8,300 at fill temperature, so one setpoint is two flow regimes and the second sits below the threshold the rule exists to stay clear of.
  4. KK-DCL-104The CDU Compatibility Table Has No Owner, So Closure Has No RecipientThe compatibility table that would reconcile the primary supply temperature against the required secondary supply temperature has no single responsible party, so the document that would close the rating is registered with a named supplier and nobody to receive it.