Amanda Prall · Design Ops + UX LeadershipAvailable for new roles
Amanda Prall

CS-02Product designVertiv Geist rack PDUEmbedded HMI + web

Nine Screens to One

Two interfaces for one device: a touchscreen the size of a postage stamp, and the browser you reach it from.

A rack power distribution unit is the strip of outlets that feeds every server in a rack. It is unglamorous and it is load-bearing: when it has a problem, everything plugged into it has a problem.

The work was two interfaces for the same device, designed in parallel. One is a 1.5 inch touchscreen physically on the unit, for the person standing in the aisle. The other is a web interface reached from a browser or a tablet, for the person managing many units from somewhere else. Both had to carry device status, error indication, and monitored values. The same information, two radically different amounts of pixel space.

Six final on-device screens: settings and alarms menus, a status menu, an exceeded threshold detail showing current value against threshold, an acknowledged alarm, and a two-alarm list, each carrying a colored status header
The final on-device screens. 128 by 128 pixels each, shown enlargedOpen full resolution ↗

The small screen is the interesting constraint. My canvas was 128 by 128 pixels. Everything a person needs to do at the device, including setup, alarm review, and acknowledgment, had to happen inside a square about the size of a postage stamp, using nothing but a fingertip, in a cold aisle, probably in a hurry.

Asking first

We started with workshops rather than prototypes. My manager and I facilitated, and the room held customers, product team members, and technicians together rather than separately. Three teams worked in parallel on landing page design, information architecture, and the commissioning flow, so the structural questions got worked through by the people who would live with the answer.

We also sat down with three external users and went through navigation, functionality, layout, interactions, and styling, marking pain points in one color and design ideas in another. That matrix is the reason the first prototype was not a guess.

Workshop output showing a proposed information architecture with grouped menu regions
Team two: information architectureOpen full resolution ↗
A dense grid of user feedback entries color coded green for design ideas and red for pain points
The user feedback matrix. Pain points in red, design ideas in green; client identifiers are not legible at this resolution, which is intentional

Design, test, change, test again

I prototyped both interfaces and wrote the test protocol. Then two rounds of usability testing, twelve sessions in total across internal staff and external customers, each observation coded against four outcomes: task validated, task not validated, user confused, or design decision required.

Device touchscreen, seven sessions in August 2021: 152 validated, 27 not validated, 29 confused, 36 design decisions.

Web interface, five sessions in October and November 2021: 206 validated, 32 not validated, 32 confused, 151 design decisions.

Roughly half of everything tested came back validated. The next largest category was not failure but decision: places where users had shown us something and the team now had to choose. Confusion and outright failure were the two smallest categories in both rounds.

Pie chart of coded usability testing observations for the device interface, dominated by the validated category
Coded observations, device touchscreen, seven sessionsOpen full resolution ↗

The constraint that arrived late

Round one of the device screen assumed scroll, swipe, and tap, borrowing the interaction model from a smartwatch because the sizes are comparable. That turned out to be a guess. Nobody had the touchscreen specification yet; Product did not have it either, and we were waiting on the manufacturer. So I designed against what similar hardware had supported in the past.

When the specification came back, scroll and swipe were not reliably available. That cost a full round of designs.

Designing forward on a stated assumption, and paying the iteration cost when it did not hold, kept the project moving through a gap that was not ours to close. What I do differently now is put the assumption in writing where the work lives: this design depends on scroll being available, and if it is not, these screens change. Then the rework is expected instead of a surprise.

Six early device screens using a scrolling list interaction model with a circular progress ring and stacked menu items
Round one. Scroll, swipe, and tap, all of which needed hardware confirmation we did not yet haveOpen full resolution ↗
Seven revised device screens using tap-only navigation with full-width solid buttons and a persistent menu control
Round two. Tap-only navigation, and a button treatment matching the existing UX standards so the device stayed consistent with the rest of the product familyOpen full resolution ↗

Making the header do the work

Round three moved status into a colored header bar tied directly to alarm state, so the condition of the device is legible before you read a single value. That round also surfaced the question that matters most on a screen this small: what information deserves size and weight, and what can be smaller. Round four added a dark variation and pushed icons and status symbols in place of words, because on 128 pixels a word costs more than a symbol.

Nine device screens introducing a colored status bar across the top of each screen in green, amber and red
Round three. Status moves into the header, so severity reads before any value doesOpen full resolution ↗
Ten device screens in a dark variation using icons and status symbols in place of words
Round four. A dark variation, and symbols where words were costing too much roomOpen full resolution ↗

What testing changed

Users told us the menu naming was confusing, and they were right. “Inlet” became “Input.” “Protection” became “Breakers.” The event log moved out of Settings and into Alarms, where people kept looking for it.

Two versions of the same status menu side by side, the first labelled Inlet and Protection, the second labelled Input and Breakers
Before and after. Two words, and people stopped hesitating

The harder finding was alarm acknowledgment. Users could not tell whether acknowledging an alarm cleared it. Working through it with the team surfaced that acknowledgment depends on user permissions, and there is no login at the device, so acknowledgment could not honestly live there at all. The resolution was to show that an alarm had been acknowledged elsewhere without offering the action locally. A design problem that turned out to be a permissions problem.

Nine screens to one

The web interface started by borrowing the framework of an existing product, which got us moving and inherited a setup process nine screens long. Testing was blunt about it: the configuration process was too complex, priority data was not prioritized, and alarm status was inconsistent.

Early web dashboard with circular gauges for input voltage and load, a status summary table, and a left navigation column
Round one. Inherited framework, gauges, and a nine-step setup behind itOpen full resolution ↗

Round two turned configuration into a wizard and added graphical elements for faster reading. Better, still not right. The final round collapsed nine screens into one, offering three configuration methods and letting the user pick the one that matched how they already worked, rather than marching everyone down a single path.

Configuration entry screen offering three methods: device interface, configure network settings, and import a file
The final configuration entry. Three methods, one screenOpen full resolution ↗
Quickstart wizard step one, create a system password, with a three-step progress indicator across the top
What used to be nine screens, now a three-step Quickstart with a visible position indicatorOpen full resolution ↗

The dashboard changed too. Gauges became bar graphs; the gauges looked good, but they made values hard to compare against one another, which is most of what a dashboard is for. Data was re-laid to put emphasis where importance was. Alarm notification was simplified down to color, clearer wording, and a count.

Final web dashboard in a normal state showing input voltage, per-phase load bars, energy totals and a status summary table
Final dashboard, normal stateOpen full resolution ↗
Final web dashboard in an alarmed state with a red exceeded alarm threshold banner across the header and three active alarms flagged
The same dashboard with an alarm running. Severity in the banner, count beside itOpen full resolution ↗

The outcome

The screens were signed off as ready for engineering, with the agreement that engineering would come back to the UX team for clarification as needed.

The device interface shipped. The rack PDU monitoring device sold in this line today carries the touchscreen I designed: the same 128 by 128 canvas, the status header tied to alarm state, tap-only navigation with menu and back sitting where the thumb already is. Some details changed in implementation. The structure and the interaction model are the ones that came out of four rounds and seven test sessions.

The browser experience carried the work forward as well, following the layout and the updated design system, and extending it the way it should: everything you can reach at the device, plus the functionality that only makes sense when you have the room for it.

A rack PDU is not a product anyone photographs for a design annual. It is a strip of outlets in a cold aisle. But somewhere right now a person is standing in front of one, tapping a square the size of a postage stamp, and finding out in a glance whether they have a problem. That is the whole job.

What I did

Designed both interfaces: the on-device touchscreen across four prototype rounds and a final, and the web interface across two rounds and a final. Co-facilitated the discovery workshops with my manager, running customers, product staff and technicians through information architecture, layout and commissioning flow in one room. Wrote the usability test protocol, and coded and analyzed twelve sessions across two rounds. Worked through the alarm acknowledgment finding with the team to its permissions root. Built to the existing UX standards so the device stayed consistent with the wider product family.

Prototypes from 2021 and early 2022, shown as they were. Customer names and site identifiers removed.