Packet loss

The share of packets that never arrive.

Definitions only — a term explained here is not a service we sell.

Packet loss is the proportion of packets sent that do not reach their destination. It can happen at a congested queue, a failing link, a radio error or a policy drop.

Why it matters

A lost packet is retransmitted or abandoned. For a bulk transfer, retransmission costs time; for a real-time stream, there is no time to retransmit, so the application hides the gap or drops quality.

Delay and loss are different failures

Delay and loss are different failures
Value Zone What it means
p50 20 ms Typical packet
p95 60 ms Slow tail
p95 − p50 40 ms Variation

Illustrative example; not a measurement of our network.

The strip shows the delay of packets that arrive. Loss is the share that never appears and needs its own counter.
Text description

The strip cannot show loss, because a lost packet has no delay to plot. Read it as the timing of the packets that did arrive, and read loss from a separate counter over the same window.

What to check

Ask for loss in both directions, at the packet size and rate you use. A loss figure from a large probe packet may not match the loss a small real-time packet sees, and an average over a day hides a peak-hour burst. The jitter term and the throughput term cover the other measurements.

FAQ

Is packet loss the same as a dropped call?

No. A dropped call is a session failure; packet loss is a per-packet event that may or may not end the session.

What causes packet loss?

Congestion, radio errors, a failing link, a full buffer or a policy drop. The cause decides whether retransmission helps.

What loss is acceptable?

It depends on the application. There is no single number; ask for the measurement method and the traffic profile.

Learn more

Sources

  1. ITU-T Y.1541: Network performance objectives for IP-based services (packet loss)
  2. ITU-T G.114: One-way transmission time