Why A Video Call Degrades But A Download Does Not

A congested connection makes a download slow and a video call ugly. Both are the same network behaving the same way, interpreted by two protocols with opposite priorities.

One protocol guarantees delivery

The transport used for files, web pages and email numbers every segment and waits for acknowledgement. Anything unacknowledged is sent again until it arrives.

Nothing is delivered to the application out of order. If segment four is missing, segments five and six wait in a buffer until the gap is filled.

The result is that data is never wrong, only late. A file downloaded over a terrible connection is identical to one downloaded over a perfect one.

The other protocol guarantees nothing

The alternative transport sends packets and forgets them. There is no acknowledgement, no retransmission and no reordering.

Packets may arrive out of sequence, duplicated or not at all, and the application is left to cope. In exchange, nothing ever waits.

For live audio and video this is the correct trade. A frame that arrives late is worthless, so there is no point spending time recovering it.

Congestion control creates the slowdown

The reliable protocol interprets loss as a sign the network is overloaded and reduces its sending rate sharply, then climbs back slowly.

This is why a download's speed oscillates rather than holding steady. It is deliberately probing for the limit and backing off each time it finds one.

Because every connection does this, they converge on a rough share of the available capacity. The mechanism is what keeps a shared network from collapsing entirely.

Real-time media handles loss itself

A call cannot back off in the same way without introducing delay, so it adapts differently by reducing resolution, frame rate or audio detail.

Lost packets are concealed. Video repeats or predicts regions of the picture, which produces the smeared blocks that appear when a call struggles.

Some systems send redundant data deliberately, accepting a higher rate so that a lost packet can be reconstructed without asking for anything.

The two interact badly on one link

A large download filling every buffer on the path adds delay to everything sharing it, including the call that cannot tolerate delay.

The download is behaving correctly by its own rules, and the call has no way to ask it to stop. Neither protocol knows the other exists.

Queue management on the router is the only place this can be fixed, by keeping buffers short and giving small time-sensitive flows priority over bulk transfers.