Data Center Colo · Case study

Data center colo — power train & SLA.

Two scenarios from a 14 MW Atlanta colo — a CRAH drift caught before a flagship tenant noticed, and a $48K power-event SLA claim resolved in five business days.

A 14 MW Atlanta colo facility with hyperscale and enterprise colo tenants. BMS averages room temperature. DCIM polls every 5 minutes. Meanwhile a single CRAH’s supply temperature climbs 2°C and a rack starts drifting into ASHRAE A1 upper. Two scenarios: one thermal, one SLA dispute.

6 minCRAH baseline restoration — no tenant exposure
$48KSLA claim withdrawn on per-outlet evidence
5 daysto dispute resolution
Operator scenarios

How this plays out in the field.

01

The drift the BMS averaged away.

ATL-DC-3 · CRAH-7 supply temp · +2.1°C above baseline · sustained 45 min
01
ATL-DC-3 · CRAH-7 supply temp · +2.1°C above baseline · 45 min

BMS averaged it away. Rack intake didn’t.

BMS reports room-level average temperature as clean. DCIM polls every 5 minutes — also fine. But CRAH-7’s supply temperature has been climbing 2.1°C above baseline for 45 minutes across three 15-minute windows. The ML inference racks it primarily cools are creeping into ASHRAE A1 upper. Tenant SLA penalty: $14K per hour if the envelope is breached.

02
correlation engine · chiller plant rebalance at 04:12 · CRAH-7 under-fed

Root cause: chiller rebalance left one CRAH under-fed.

Cooling pillar correlates CRAH-7’s supply drift against the chiller plant load-balance log. Root cause: a plant rebalance at 04:12 left CRAH-7 under-fed. Intervention runbook attached: two valve adjustments, estimated 10 minutes to execute. Forecast envelope remains inside ASHRAE A1 if corrected within 20 minutes.

03
EN 50600 · operations log sealed · ASHRAE A1 maintained

Corrected in 6 minutes. Tenant never notices.

Operator approves the runbook. Valve adjustments executed. CRAH-7 returns to baseline within 6 minutes. Two ML inference racks never exceed the A1 envelope. EN 50600 operations log entry sealed automatically. Q3 facility report adds one more entry to the “caught and resolved before tenant exposure” column.

6 min
to baseline restoration
From runbook approval to CRAH-7 supply temperature back within ASHRAE A1 envelope.
A1
ASHRAE envelope — maintained
ML inference racks never exceeded class A1 intake temperature ceiling. No tenant exposure.
ASHRAE TC 9.9EN 50600
02

The tenant’s monitoring said one thing. Yours said another. Yours was sealed.

DAL-DC-1 · tenant T-09 · SLA dispute · 22-min power event · cabinet C-44
01
DAL-DC-1 · tenant T-09 · SLA dispute · 22-min power event · cabinet C-44

Tenant claims 22 minutes of power exposure.

A colo tenant files a quarterly SLA dispute claiming a 22-minute power event in cabinet C-44. Internal monitoring shows a server reboot loop in the window. Facility BMS shows the room-level UPS feed as clean. Dispute claim: $48K in SLA credit. Tenant renewal in four months.

02
Power pillar · per-outlet continuity · C-44 · ±0.4% · sealed

Per-outlet continuity: continuous within ±0.4%.

ObservOne’s Power pillar tracked power from utility feed to per-outlet on each PDU feeding C-44. Hash-chained log captured every reading: utility feed steady, ATS not exercised, UPS string steady, cabinet outlets delivering continuous power within ±0.4% across the entire disputed window.

03
dispute packet · sealed · claim withdrawn in 5 business days

$48K claim withdrawn. Kernel panic, not facility power.

Compliance pillar drafts the dispute response: per-outlet continuity timeline for C-44, sealed against the operator’s tenant key, raw event stream attached. Tenant’s IT team review. Their server reboots traced to an internal kernel panic on a cluster patch. Claim withdrawn in five business days.

$48K
SLA claim withdrawn
Per-outlet continuity evidence resolved what a BMS screenshot could not.
5 days
to dispute resolution
Tenant IT team traced reboots to a kernel panic, not facility power.
SOC 2EN 50600Uptime Institute Tier
Hands on

Reproduce this scenario
in our sandbox.

30 minutes with a solutions engineer. We'll preload a tenant with anonymised sites matching your topology. NDA-friendly.