Delay, loss, and throughput
How fast a network feels depends on several kinds of delay, on lost packets, and on the throughput of its slowest link. These measures explain everyday performance.
Four sources of delay
At each hop, a packet experiences processing delay (examining headers and choosing a link, usually microseconds), queueing delay (waiting behind other packets, from zero to many milliseconds), transmission delay (pushing the packet's bits onto the link), and propagation delay (the time for a signal to travel the link's length). Total nodal delay is their sum.
Transmission versus propagation
Transmission delay is L / R: packet length L bits divided by link rate R bits per second. A 12,000-bit packet on a 100 Mbit/s link takes 120 microseconds. Propagation delay is d / s: distance divided by signal speed, about 2 x 10^8 metres per second in fibre, roughly two-thirds the speed of light in a vacuum. A 2,000-kilometre fibre adds about 10 milliseconds, whatever the link rate.
Queueing and traffic intensity
If packets arrive at an average rate a packets per second, each L bits, on a link of rate R, the traffic intensity is La / R. As it approaches 1, average queueing delay grows sharply; above 1, the queue grows without limit. Even below 1, bursts cause temporary queues. Router buffers that are too large can cause excessive delay, a problem called bufferbloat.
Packet loss
Buffers are finite. When a packet arrives at a full queue, it is dropped. Losses may be recovered by retransmission at the transport layer, as TCP does, or simply tolerated, as in many live video and voice applications. Rising loss and delay are the main symptoms of congestion.
Throughput and bottlenecks
Throughput is the rate at which data actually arrives at the receiver. For a path of several links, end-to-end throughput is limited by the slowest one, the bottleneck link. Today, the bottleneck is usually in the access network, such as home Wi-Fi or a mobile connection, rather than the core, though shared links split their capacity among many flows.
Round-trip time and bandwidth-delay product
Round-trip time (RTT) is the time for a small packet to travel to a destination and back. The bandwidth-delay product, link rate times RTT, gives how much data can be 'in flight' on the path. A 1 Gbit/s path with an RTT of 100 milliseconds holds 100 megabits, about 12.5 megabytes, in flight, which matters for how transport protocols must be tuned.
Measuring performance
The ping tool measures RTT using ICMP echo messages; traceroute reveals each router hop and its delay by sending packets with increasing time-to-live values; speed tests measure throughput. Latency, jitter (variation in delay), loss, and throughput together determine the quality of applications from file downloads to video calls and online games.
Worked example
Sending a 1,500-byte (12,000-bit) packet across 3 links, each 10 Mbit/s and 1,000 kilometres long, ignoring processing and queueing: transmission per link is 12,000 / 10,000,000 = 1.2 milliseconds; propagation per link is 1,000,000 / (2 x 10^8) = 5 milliseconds. With store-and-forward at each hop, total delay is 3 x (1.2 + 5) = 18.6 milliseconds.
Key takeaways
Nodal delay sums processing, queueing, transmission (L / R), and propagation (d / s) delays. Signals travel about 2 x 10^8 m/s in fibre, so distance sets a minimum latency. Traffic intensity La / R near 1 causes sharply rising queueing delay; oversized buffers cause bufferbloat. Full buffers drop packets, signalling congestion. End-to-end throughput is set by the bottleneck link, usually the access network. RTT and the bandwidth-delay product describe how much data is in flight. Ping, traceroute, and speed tests measure latency, hops, and throughput; jitter and loss matter for real-time apps.
Flashcard
Examining headers and choosing a link
Check your understanding
Upgrading a 1,000-kilometre fibre link from 1 Gbit/s to 10 Gbit/s barely reduces the delay of small packets. Why?