Your upload speed is the number that decides how the connection feels

Everything here earned its place by changing an outcome. Nothing about upload bandwidth is included to round the number up.

Asymmetry is a design decision

Cable and DSL technologies divide a shared frequency range between downstream and upstream, and operators allocated most of it downstream because that matched consumption. Fibre to the premises has no such constraint physically, yet many providers still sell asymmetric packages by policy. The assumption was that homes consume far more than they produce, which was true until video calling and cloud backup became routine.

The ratio, not just the headline figure, is what determines how a connection behaves under real household load.

A full uplink slows your downloads

Reliable transport protocols confirm received data with small acknowledgement packets travelling in the opposite direction. If the uplink is saturated, those acknowledgements are delayed or dropped, and the sender slows down as though the network were congested. This is why starting a large upload can make an unrelated download collapse, which looks illogical and is entirely expected.

Mechanically, prioritising small packets in the router mitigates it substantially.

Video calls are upstream-heavy by nature

Your outgoing video and audio consume uplink continuously, and unlike a download they cannot be buffered ahead. When uplink is constrained, the software reduces resolution and frame rate, which is why you look worse to others than they do to you.

Mechanically, several people on calls in the same house multiply the requirement, and it is the least tolerant traffic on the network. Packet loss upstream degrades a call far more visibly than a shortfall in raw capacity.

Backup and sync compete quietly

Cloud backup and photo sync upload continuously in the background and will use whatever uplink they find. Most such software offers a bandwidth limit, and setting it to a fraction of your uplink is one of the highest-value network settings available. Scheduling large uploads outside working hours achieves the same result more crudely.

A home camera streaming to the cloud does this permanently rather than occasionally.

Bufferbloat makes it worse than it should be

Oversized buffers in routers and modems hold packets rather than dropping them, which delays the congestion signal transport protocols rely on. The result is latency rising into the hundreds of milliseconds while a transfer runs, even though throughput looks fine. Modern queue management algorithms address this directly and are available in many routers, sometimes under a name referencing smart queueing.

Enabling them and setting the shaping rate slightly below your actual line rate is the standard fix.

What to measure

A speed test run while the line is idle tells you the ceiling; a latency test run while the line is loaded tells you how it will feel. The difference between idle and loaded latency is the single most useful broadband measurement most people never take.

Several public test tools report it explicitly, and a well-configured connection shows only a small rise. A connection with modest headline figures and low loaded latency beats a fast one that stalls under load.

Everything above, in order of what to do first

1. Asymmetry is a design decision. Cable and DSL technologies divide a shared frequency range between downstream and upstream, and operators allocated most of it downstream because that matched consumption.
2. A full uplink slows your downloads. Reliable transport protocols confirm received data with small acknowledgement packets travelling in the opposite direction.
3. Video calls are upstream-heavy by nature. Your outgoing video and audio consume uplink continuously, and unlike a download they cannot be buffered ahead.
4. Backup and sync compete quietly. Cloud backup and photo sync upload continuously in the background and will use whatever uplink they find.
5. Bufferbloat makes it worse than it should be. Oversized buffers in routers and modems hold packets rather than dropping them, which delays the congestion signal transport protocols rely on.
6. What to measure. A speed test run while the line is idle tells you the ceiling; a latency test run while the line is loaded tells you how it will feel.

The takeaway

The download figure sells the package. The upload figure decides whether the call holds up.

Understanding the failure mode tells you more than the feature list does.