Count the Site Visits Before Counting the IOPS
The economics of an edge node differ from a central facility: sites are scattered, each is small, and none has an engineer on hand. One physical visit costs more than the price gap across several drives. The first filter in selection is therefore not a performance figure but whether a fault can be handled without dispatching anyone. Put that first and the later discussion of bandwidth, IOPS and endurance at least has a boundary. The portfolio's 30 formal models are all at the product-definition/EVT stage and every performance and power number is a planning target — yet the feature baseline list is what gets used first in this scenario, ahead of any single performance point.
Power Loss, Logs, Upgrades: Three Things That Set the Service Radius
Sites with unreliable mains see unexpected outages every year. Power-loss protection decides whether an outage forces a manual consistency check. End-to-end protection and T10 DIF/DIX decide whether an anomaly can be pinned to a specific stage. NVMe Telemetry and NVMe-MI allow fault logs and out-of-band management data to be pulled centrally, avoiding the situation where a serial console on site is the only option. Online firmware update with rollback determines the consequence of a failed batch upgrade — with rollback it is one remote command, without it, a drive out. Multiple namespaces and TCG OPAL/Secure Boot/Sanitize sit in the same baseline, while SR-IOV and OCP-related capabilities are scoped to phase P1 and should not be assumed present when selecting for the edge.
How Many Nodes Fit a Rack: Maximum Power Sets the Ceiling, Idle Power Sets the Bill
Edge facilities generally have weaker power and cooling than a central data centre. The CORE 7 J750 and J770 plan for 18W typical, 22W maximum and 6W idle draw, and across a multi-bay node those numbers multiply by twenty-something. Rack budgets must be built from maximum draw; treating the typical figure as an average walks straight into the supply ceiling during traffic peaks. The 6W idle figure, meanwhile, is present continuously through low-traffic overnight hours and is the real driver of the electricity bill. The two numbers govern two different things and cannot stand in for each other.
The Edge Usually Means Reusing Chassis, Not Deploying Platforms
Edge sites rarely swap out an entire platform at once; replacing drives in existing machines is far more common, and that alone bounds the candidate list. On Gen3 platforms, LEGACY 3's J360 and J380 cover installed servers and industry-specific machines, while X3's J390X addresses low-latency replacement in the installed base. Only machines with Gen4 slots bring CORE 7's J750 and J770 into play, and those two give a choice between 1 DWPD and 3 DWPD, with the latter suiting nodes that refill from origin frequently. Note also that read and write peaks at the edge tend to coincide — when hit rate drops, origin-refill writes land right on top of the read peak, so behaviour under mixed load deserves more scrutiny than any single headline figure.