Skip to content
Tutorials Center
My learning

Topics

Lesson 4 of 79 ยท Getting Started with Networks

Bandwidth, Latency, Throughput and Loss

Learn the four delays a packet suffers, bandwidth versus throughput, the bandwidth-delay product, RTT, jitter and packet loss, with live delay calculators.

"My Internet is slow" can mean many things. The download may start late, a video call may stutter, a game may lag, or a page may load in pieces. These are different problems with different causes, and network engineers use precise words to tell them apart: bandwidth, throughput, latency, jitter and loss.

This lesson defines each term, shows where delay comes from (four separate sources), introduces the bandwidth-delay product, which explains why long, fast links are hard to fill, and lets you compute everything yourself. It follows How Data Travels and prepares the ideas behind TCP flow control and congestion control.

Bandwidth and throughput

Bandwidth (here meaning link capacity) is the maximum number of bits a link can carry per second, written in bit/s, kbit/s, Mbit/s or Gbit/s. A 100 Mbit/s Ethernet link can push at most 100 million bits onto the wire each second. Network speeds use bits; file sizes use bytes, and a byte is 8 bits, so 100 Mbit/s is at most 12.5 MB/s.

Throughput is the rate at which data is actually delivered between two hosts over a period of time. It is always lower than or equal to the bandwidth of the slowest link on the path, because of protocol headers, waiting in queues, retransmissions, and competition from other traffic. The slowest link on a path is the bottleneck, and it limits the throughput of the whole path no matter how fast the other links are.

A useful analogy is a water pipe: bandwidth is the width, throughput is the water that actually comes out per second, and latency is how long the first drop takes to arrive.

The four sources of delay

A packet moving across a link and through a router suffers four delays at each hop. The total delay at a hop is their sum.

  1. Processing delay. The router reads the header, checks for errors and picks the output link. Usually microseconds.
  2. Queuing delay. The packet waits in a buffer until the output link is free. It ranges from zero on an empty link to many milliseconds (or lost packets) when congested. This is the only delay that varies a lot.
  3. Transmission delay. The time to push all the bits of the packet onto the link: L / R, where L is the packet length in bits and R the link rate in bit/s. A 1,500-byte packet on a 100 Mbit/s link needs 0.12 ms.
  4. Propagation delay. The time for a bit to travel along the medium: d / s, where d is the distance and s the propagation speed. In fiber and copper s is about 2 ร— 10โธ m/s, roughly two-thirds of the speed of light in a vacuum.
flowchart LR
  A["Packet arrives<br/>at router"] --> B["Processing<br/>read header, choose link"]
  B --> C["Queuing<br/>wait for the link"]
  C --> D["Transmission<br/>L / R, push bits out"]
  D --> E["Propagation<br/>d / s, travel the wire"]
  E --> F["Arrives at<br/>next device"]

Four delays at every hop; only queuing changes much from moment to moment.

Transmission and propagation are often confused. Transmission delay depends on the packet size and link rate; making the link faster shrinks it. Propagation delay depends on distance; making the link faster does nothing to it. That is why a 10 Gbit/s link between continents still has tens of milliseconds of delay: light needs time to cross the ocean.

Python Delay calculator

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

On the office cable, transmission (0.12 ms) dominates and propagation is negligible (0.0003 ms). On the long fiber link the picture flips: transmission is only 0.012 ms, but propagation is 20 ms. Upgrading the long link to 10 Gbit/s would barely change the total; moving the server closer would.

Round-trip time

Most conversations need a reply, so what matters is the round-trip time (RTT): the time from sending a packet until the answer returns. Ignoring queues, it is about twice the one-way delay. The ping tool measures RTT by sending a small message and timing the response (see Essential Network Tools).

RTT matters for every protocol that waits for an answer before continuing. Opening a connection, asking for a web page, and requesting a DNS name each cost at least one RTT. Three sequential steps over a 100 ms RTT path take 300 ms before a single byte of the page arrives. Many performance tricks (keeping connections open, caching, placing content close to users) are really attempts to cut the number of round trips.

The bandwidth-delay product

How much data is "in flight" on a link at one instant? Multiply the rate by the one-way delay (or, more usefully, by the RTT):

BDP = bandwidth ร— RTT

The bandwidth-delay product is the amount of data a sender must have sent, but not yet seen acknowledged, to keep the link full. Think of the pipe again: its volume is width times length. If the sender stops after sending less than the BDP and waits for an acknowledgment, the pipe sits partly empty.

This is why window-based protocols such as TCP can be limited by their window size instead of by the link. If the sender may only have W bits outstanding, its throughput cannot exceed W / RTT, whatever the bandwidth. The next example computes both numbers.

Python Bandwidth-delay product and window limits

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

To fill a 100 Mbit/s path with a 150 ms RTT you need 1,875 kB in flight. A sender restricted to a 64 KiB window can reach 100 Mbit/s only when the RTT is about 5 ms; at 150 ms it manages just 3.5 Mbit/s, on a link that could carry 100. This is the reason TCP added window scaling, covered in TCP Flow Control.

Jitter and loss

Jitter is the variation of delay from packet to packet. A steady 40 ms is fine for a video call; a delay that jumps between 20 ms and 50 ms is not, because audio packets arrive in uneven bursts. Applications hide jitter with a playout buffer that holds incoming packets briefly and releases them at an even pace, trading a little extra latency for smoothness. In the sample above the mean RTT is 25.6 ms, but two outliers (35.2 and 48.9 ms) push the standard deviation to 10.1 ms, and the simple jitter figure (average change between consecutive samples) is 12.7 ms. Mean latency alone would hide this.

Packet loss is the fraction of packets that never arrive. Reasons include full queues (congestion), corrupted bits that fail a check, a failed link or a busy wireless channel. In the example, 12 of 500 packets were lost: 2.4%. Loss hurts differently depending on the application. A file transfer simply retransmits the missing pieces, which costs time; a live call usually skips them, which costs quality. TCP also treats loss as a signal to slow down, so even 1% loss can cut the throughput of a long-distance transfer dramatically.

Latency, bandwidth and what to improve

Match the remedy to the symptom.

Symptom Likely culprit Typical remedy
Big downloads take forever Low bandwidth or a bottleneck link Faster plan, fix the slow hop
Pages feel sluggish even on a fast plan High RTT times many round trips Fewer requests, nearby servers, caching
Calls stutter Jitter or loss Wired connection, smaller buffers on the router, less competing traffic
Speed varies by time of day Queuing from congestion Off-peak use, quality-of-service rules

What is real and what is a stand-in? The formulas are the standard textbook models and the numbers follow from them exactly. Real paths have many hops, each with its own queue, and measured RTTs include operating-system delays that the model leaves out. The RTT samples are invented to illustrate jitter.

Check your understanding

Quick check ยท 5 questions

of 5 answered You scored / 5 ยท perfect!

  1. 1. A 12,000-bit packet is sent on a 1 Mbit/s link. What is the transmission delay?

    L / R = 12,000 / 1,000,000 = 0.012 s = 12 ms.
  2. 2. Which delay does NOT shrink when you replace a link with a faster one of the same length?

    Propagation delay depends on distance and signal speed, not on the link rate. Only transmission delay depends on the rate.
  3. 3. A path has 80 Mbit/s bandwidth and a 50 ms RTT. What is its bandwidth-delay product?

    80,000,000 bit/s ร— 0.05 s = 4,000,000 bits, which is 500 kB.
  4. 4. What is jitter?

    Jitter is the variation of delay from packet to packet, which matters most for real-time audio and video.
  5. 5. A sender may have at most 128 KiB outstanding and the RTT is 100 ms. What is the throughput limit?

    W / RTT = 128 ร— 1024 ร— 8 bits / 0.1 s = 10.49 Mbit/s, whatever the link could carry.

Summary

  • Bandwidth is link capacity in bits per second; throughput is the rate actually delivered, limited by the bottleneck link.
  • Each hop adds processing, queuing, transmission (L / R) and propagation (d / s) delay; only queuing varies much.
  • RTT is the delay of a round trip; many protocols pay it several times before data flows.
  • The bandwidth-delay product (bandwidth ร— RTT) is the data needed in flight to fill a link; a window smaller than that caps throughput at W / RTT.
  • Jitter is delay variation and loss is the fraction of missing packets; real-time applications are sensitive to both.

Next: Layered Models: OSI and TCP/IP.