Enough to answer the question you will eventually be asked: what is on this port, and why is it slow.

Unmanaged switches are genuinely fine in places where the network is flat, small and unimportant - a few devices behind a desk. The moment a site has more than one class of traffic, or anybody is expected to diagnose a fault remotely, the absence of management stops being a saving.

The practical dividing line is visibility. A managed switch tells you port status, error counters, traffic rates, what MAC addresses it has learned and where a loop is. An unmanaged one tells you nothing, so every fault becomes a physical process of unplugging things.

The second line is segmentation. Once cameras, building controls, guest devices and business systems share a cabling system, VLANs are the mechanism that keeps them apart, and that requires managed switches end to end.

Buying unmanaged and replacing later is expensive, because the replacement happens during an outage.

The count you need today plus growth, but more importantly the uplink capacity behind them and the space and power in the cabinet.

Port count itself is the easy part: count the outlets, add the devices that will arrive, and leave headroom. The usual guidance of a quarter to a third spare is not generous - it is what stops the next small project needing a new switch.

What is missed more often is the uplink. Forty eight ports at a gigabit each feeding a single gigabit uplink is a bottleneck by design, and it will not show as an error, only as slowness at busy times. Uplinks should be sized against realistic aggregate demand, and are usually fibre.

Also check the cabinet. Two 24-port switches and one 48-port switch fill different amounts of rack space, need different power, and fail differently - the 48-port unit takes twice as many users down with it.

A Layer 2 switch moves traffic within a network segment. A Layer 3 switch also moves traffic between segments, in hardware, without sending it somewhere else to be routed.

On a Layer 2 design, each VLAN is a separate broadcast domain and anything that has to cross between VLANs is sent up to a router or firewall, which routes it and sends it back down the same link. That is called hairpinning, and it consumes uplink bandwidth twice for every packet.

A Layer 3 switch has a routing engine in silicon, so inter-VLAN traffic is routed at wire speed inside the switch and never leaves it.

The decision is usually about volume and about where policy is enforced. If most traffic stays within its own VLAN, Layer 2 at the edge is fine. If VLANs talk to each other constantly, or the uplink is congested by traffic that never needed to leave, Layer 3 belongs at the aggregation point.

By adding the power every connected device draws, then checking that figure against the switch's total budget rather than its per-port rating.

Every PoE switch quotes two numbers. The per-port figure is the maximum a single port can deliver, set by the PoE standard it supports. The total budget is what the internal power supply can deliver across all ports at once, and it is almost always less than the per-port maximum multiplied by the port count.

So a 24-port switch that can deliver 30 W to any port may only have a 370 W budget - comfortably enough for phones, not enough for 24 heated outdoor cameras.

Add up the actual devices, including their peak draw rather than idle, and add the cable loss over long runs. Then leave headroom, because the devices added later are usually the hungry ones.

When the budget runs out the switch stops powering ports by priority, which is a confusing fault if nobody expected it.

No for the basics, and increasingly yes for the features people actually buy switches for.

Ethernet, VLAN tagging, link aggregation and spanning tree are standards, and switches from different manufacturers interoperate on all of them. A mixed estate works at that level.

Where it stops being simple is everything above: stacking, controller-based management, single-pane monitoring, automated configuration, and the proprietary fast-convergence protocols used in industrial rings. None of those cross manufacturers.

So the practical answer depends on how the estate is run. A site managed device by device can mix freely. A site that expects one management interface, consistent configuration templates and firmware managed centrally should standardise, because the operational cost of not doing so exceeds the hardware difference.

Mixing at a boundary - one manufacturer in the data centre, another at the edge - is common and works, provided the link between them uses only standards.

Two paths between the same two switches with nothing to block one of them - and it is serious because Ethernet has no mechanism to stop a frame circulating.

An IP packet has a time-to-live that expires. An Ethernet frame does not. A broadcast frame that finds a loop is copied endlessly around it, multiplying at every switch, until the links are saturated and every device on that segment is unable to communicate. It happens within seconds.

Loops are almost always created accidentally: a patch lead plugged into two outlets, a small unmanaged switch someone added under a desk, a second uplink patched in during a change.

Spanning tree protocol exists to prevent it, by detecting redundant paths and blocking them until they are needed. It only works if the switches are managed and it is enabled - which is the strongest single argument against unmanaged switches in a shared cabling system.

Edge protection features such as loop detection and BPDU guard shut the offending port instead of the network.

Airflow direction, power, earthing and the cable management - in that order, because all four are difficult to change afterwards.

Switches move air either side to side or front to back, and mixing directions in one cabinet makes one switch inhale another's exhaust. Check the direction against the cabinet's own airflow before mounting.

Power needs to be counted honestly, including PoE draw, and split across feeds if the switches have dual supplies - dual supplies on one power strip provide no redundancy at all.

Earthing and bonding of the cabinet and the patch panels matters for screened cabling systems, and is frequently forgotten in retrofits.

Finally, leave room for patch leads. A 48-port switch generates a substantial bundle, and a cabinet with no horizontal management and no slack ends up with leads bent below their radius and ports that cannot be reached without disturbing others.