Power

Colo power: contracted kW and A/B feeds

In colo, the provider designs, runs, and maintains utility service, UPS, generators, and distribution. You contract capacity at the cabinet or cage, then supply and manage the rack PDUs. Provider selection and general contract terms are on Facility: Colo evaluation; cage layout is on Racks: Colo fit-out. Contract rules are provider-specific and change. One provider’s published schedule is used below as an example, not a norm. Sources checked 5 Oct 2026.

What you contract

• Capacity in kW or kVA per cabinet or cage, delivered as circuits of a stated voltage, amperage, and phase
• Billing model: a flat charge per circuit, metered energy, or a mix
• Receptacle type: your rack PDU input plug must match it, so confirm before ordering PDUs
• Power readings: whether the provider offers a portal or API, and how often it updates. Your own metered PDUs are the per-circuit view you control

Size A/B on the surviving feed

When one feed of an A/B pair fails, the other carries the whole load. The pair’s combined draw must therefore fit within one circuit’s usable limit. eStruxture’s July 2026 colocation schedule caps use at 80% of fuse rating and, for an A+B pair, at 40% of fuse rating on each circuit. A customer who draws more can be billed for the circuits as standalone, non-redundant power (eStruxture schedule). Under that rule, each circuit of a pair of 30 A circuits carries at most 12 A in normal running. Equinix states a similar rule in which the usable percentage varies by region and voltage, and overdraw voids the redundant-pair SLA. Read your contract for the exact percentage, and alarm on per-circuit draw from your own PDUs.

What to check in a provider’s power SLA

• Availability by redundancy level. eStruxture example: 100% for A+B high availability, 99.99% for a standard single circuit, with a tiered credit table (same source)
• What counts as an outage (one circuit or both of a pair), and whether the SLA applies only if you stayed within the loading rule
• Credit caps and how to claim
• Whether A and B come from diverse UPS systems and diverse distribution paths
• Maintenance notice periods for the power chain
• Generator fuel autonomy on site and refuel contracts
• Generator rating class. Uptime counts continuous, derated prime, or unlimited-runtime standby ratings toward Tier III/IV (Uptime ATD paper, abstract)
• Evidence behind any rating claim. What “certified” vs “designed to” means is on Facility: Certified vs designed

Power networking: PDU, UPS, and EPMS connectivity

This section lists the network decisions that power devices create. Rack-side devices go on the out-of-band network and facility plant in the OT power zone. The split and the conduit rules are on Security: PDU and UPS cards. The EPMS platform that collects the readings is on Ops: EPMS. Sources checked 5 Oct 2026.

What needs a management port

UPS network cards, rack PDU controllers, rack transfer switches, switchgear and power meters, generator and paralleling controllers, and battery monitoring. Common protocols are SNMP, HTTPS and SSH, Modbus TCP, and BACnet/IP. Raritan documents SNMP and an open API for DCIM/BMS on PX4 (Raritan PX4 FAQ). DMTF Redfish added power distribution schemas for rack and floor PDUs, transfer switches, and switchgear in release 2019.4. At that release a UPS schema was still listed as future work (DMTF overview).

Ports, addresses, and chains

Daisy-chaining PDUs reduces both switch ports and IP addresses. Raritan PX4 cascades up to 32 PDUs. Its “port forwarding” mode shares one IP and gives each PDU its own port; “bridging” gives each PDU its own IP (Raritan FAQ above). Vertiv PowerIT states daisy-chaining of up to 50 devices on one IP address, with RSTP for fault tolerance (Vertiv PowerIT page).

A chain is a shared failure domain for monitoring and remote outlet control, but not for power delivery. PDU controllers usually draw power from their own feed. An A-feed outage or a fault on the primary PDU therefore cuts off monitoring and control of every PDU behind it. A firmware upgrade or reboot of the primary, a broken cascade cable, or a failed uplink port has the same effect on the whole chain. Some Raritan link-port cascades can share controller power; treat that as a per-model feature to check, not a default.

• Always keep A-feed and B-feed PDUs on separate chains. Where there are two OOB switches, land each chain on a different one
• If a chain is closed into an RSTP ring (Vertiv), the OOB switch ports it lands on must take part in spanning tree. BPDU guard or edge-port settings will shut those ports down. Port design is on Network: OOB
• In Raritan port-forwarding mode (one IP, a port per PDU), collector and firewall rules must allow the per-PDU ports

Decisions to make

• Which zone each device goes in: OOB or facility OT. Use the split on Security: PDU and UPS cards
• OOB switch ports and IP addresses per rack, after chaining. Use static addressing (see Network: OOB). Use DHCP only if the DHCP server sits inside the OOB itself
• Whether to allow or disable vendor cloud connectors on UPS and PDU cards
• Management protocols: SNMPv3 authPriv only, and disable SNMPv1/v2c, Telnet, and HTTP. Disable Modbus TCP and BACnet/IP on OOB rack devices unless the collector needs them, because neither protocol authenticates. Read-only enforcement, including for Redfish, is on Security: PDU and UPS cards
• Collector access and alarm destinations: traps and syslog go to the DMZ collector, and write actions come only from OOB admin sessions. See Security: PDU and UPS cards
• AAA (RADIUS or LDAP, with a local fallback account) and NTP, both reachable with no dependency on production. See Network: OOB. Accurate time matters because transfer and alarm timestamps drive root-cause analysis
• Firmware patch cadence for cards and controllers. CISA advisory ICSA-21-313-01 covered vulnerabilities in Schneider NMC2/NMC3 cards and advises keeping such devices off the Internet and behind firewalls (CISA). Schedule upgrades of a chain’s primary PDU as a monitoring outage for the whole chain
• The UPS shutdown-signaling path to hypervisors. The brokered flow and its firewall rule are on Security: PDU and UPS cards; UPS runtime is under UPS
• Colo: your PDUs need an OOB path inside the cage. Provider UPS and EPMS data comes only through the provider’s portal or API, if one is offered