When every remote path has failed, during physical work at the rack, and at commissioning - which is not often, and that is the point.

The primary case is total loss of remote access: the management network is down, the console server is unreachable, and somebody has to go and look. In that situation the tray is what makes the visit productive rather than requiring a monitor and keyboard to be found.

Commissioning is the second: initial setup of equipment before the management network exists.

Physical maintenance is the third - somebody replacing a component or reseating hardware often needs to see the console while doing it.

It is also the fallback when the KVM system itself has failed, provided the tray can be connected directly.

Because it is used rarely, it is often found not working when it is needed. Testing it periodically, as part of routine rack inspection, is a small effort that repays itself the first time it matters.

Adequate rather than good - the height constraint limits what is possible, and the specification should be checked against what will actually be displayed.

A screen that folds into one rack unit is necessarily short and wide. Resolutions vary considerably between products, and a low-resolution panel makes graphical firmware interfaces and modern operating system installers awkward to use.

Check the native resolution against what the connected equipment outputs. A machine outputting a high-resolution signal to a lower-resolution panel is either scaled - which can be unreadable for small text - or clipped.

Viewing angle matters more than it sounds, because the panel is at waist height and viewed from above.

Brightness matters in a well-lit equipment room.

Larger screens are available in taller units - some occupy two rack units - and where the tray will be used for real work rather than emergencies, the extra unit is usually worth it.

Usually yes - an integrated switch is what makes one rack unit serve the whole cabinet, and the combined product is generally cheaper than separate items.

A tray with an integrated eight or sixteen port KVM connects to every machine in the rack, and the operator selects which to display from the tray itself.

The alternative - a tray with a single connection, moved between machines by unplugging - is workable for a rack with two or three devices and tedious beyond that.

The integrated option also saves a rack unit and a power outlet compared with a separate tray and switch.

What to check is the port count against the rack's device count with room for growth, the cable or module type required per machine and its cost, and whether the tray can also be connected directly to a machine bypassing the switch - which matters when the switch itself is the problem.

Also confirm the integrated switch supports the resolution the tray's panel provides.

Rail depth matching the cabinet, plus clearance to extend fully with the door open and room for a person to stand - and the last of these is regularly overlooked.

Trays come with adjustable rails covering a range of cabinet depths. Confirm the range covers the actual cabinet, including any allowance for cable management at the rear.

Extending the tray requires the front door to be open and nothing protruding from the units immediately above and below - cable bundles from a switch mounted directly above are a common obstruction.

The space in front matters. In a data centre with contained aisles and doors at both ends, the aisle may be too narrow to stand comfortably at an extended tray, and in a hot aisle it may be genuinely unpleasant.

Mounting height is worth thinking about: a tray near the bottom of a cabinet requires kneeling, and one near the top cannot be seen. Somewhere around waist height is right, which conflicts with the instinct to put it wherever there is a spare unit.

Yes, and the layout should be specified explicitly rather than accepted as default.

Trays are supplied with various national keyboard layouts. A tray with a layout unfamiliar to the people using it is a genuine hindrance during an incident, particularly for characters used in passwords and configuration - the symbols move between layouts and are exactly what gets typed wrongly under pressure.

The connected machine's own keyboard configuration must also match, or the characters produced will not be those printed on the keys.

Touchpad or trackball is the other choice. Touchpads are more familiar; trackballs are easier to use accurately at an awkward height and are more robust.

Some trays offer a USB port on the front so a familiar keyboard or a storage device can be connected, which is genuinely useful and worth looking for.

Specify layout on the order. Correcting it afterwards means replacing the tray.

Yes - a small amount, and it should be on the same protected supply as the rack's equipment.

The screen and the integrated switch need power. Consumption is low, but the outlet has to exist and be accounted for in the rack's power planning.

Put it on protected power. A console tray that is off during a power event is useless at exactly the moment somebody is standing in front of the rack trying to understand what is happening.

Some trays draw power from the KVM cables or from a connected machine's USB, which removes the outlet requirement but ties the tray's availability to those machines.

Check the power connector and lead length - a tray mounted mid-cabinet needs its lead to reach the power distribution unit, and short leads are a common annoyance.

Include the tray in whatever routine testing covers the rack, since power supply failures on rarely-used equipment go unnoticed indefinitely.

Less often, but it remains worth having in each rack or each row - because every remote path has a failure mode.

Management controllers, IP KVM and console servers cover almost everything remotely, and a well-designed out-of-band arrangement with a cellular fallback covers most site-level failures too.

What none of them covers is the case where the management infrastructure itself has failed, or where somebody is at the rack doing physical work and needs to see a screen.

A shared tray on a wheeled trolley, or one tray per row rather than per rack, is a reasonable compromise that keeps the capability without consuming a rack unit in every cabinet.

What matters more than the specific arrangement is that the capability exists and is known to work. A data centre where the only way to get a screen onto a machine is to find a monitor and a trolley somewhere in the building has a gap that will be discovered at the worst possible time.