Chapter 10 of 48

Network Delays & Channel Capacity

When you click a link, the page does not appear instantly. Even on a "fast" connection, every packet spends some time getting pushed onto the wire, travelling across cables, waiting inside routers, and being checked by each device on the way. Add all of that up and you get the delay you feel as lag.

There is also a second question: how much data can a channel carry per second at the very most? That limit is called the channel capacity, and two famous formulas (Nyquist and Shannon) tell us what it is.

In this chapter we will learn both ideas, with formulas and worked numerical examples. These are among the most common numerical questions in university exams, GATE and placement tests.

Sender ──[push bits out]──[travel]──[wait in queue]──[process]──→ Receiver
          transmission    propagation   queuing       processing

Bits, Bytes and Units First

Most mistakes in delay problems are unit mistakes, so let's fix the units before we start.

  • Data rates (bandwidth) are measured in bits per second: 1 kbps = 1,000 bps, 1 Mbps = 1,000,000 bps, 1 Gbps = 109 bps.

  • Packet and file sizes are often given in bytes. 1 byte = 8 bits. Always convert bytes to bits before dividing by a rate.

  • Time units: 1 ms = 10-3 s, 1 µs = 10-6 s.

Rule of thumb: convert everything to bits, seconds and metres first. Then plug into the formula.

Some exam questions use 1 KB = 1024 bytes for file sizes. If the question does not say, state the assumption you are using.


The Four Types of Delay

When a packet passes through one link and arrives at the next device (a router, a switch or the destination), it experiences four kinds of delay.

Delay

What causes it

Formula / depends on

Transmission delay

Time to push all the bits of the packet onto the link

Tt = L / R (packet size / link rate)

Propagation delay

Time for one bit to travel from one end of the link to the other

Tp = d / s (distance / signal speed)

Queuing delay

Time the packet waits in a router's buffer behind other packets

Depends on traffic load; varies from packet to packet

Processing delay

Time for a device to examine the header, check for errors and decide where to send the packet

Depends on the device; usually microseconds or less

The total delay at one hop (often called the nodal delay) is:

Nodal delay = Processing + Queuing + Transmission + Propagation

1. Transmission Delay

A link can only put a fixed number of bits per second onto the wire. If the packet has L bits and the link rate is R bits per second, then pushing the whole packet out takes:

Transmission delay  Tt = L / R

Example: A 1,500-byte packet on a 10 Mbps link.

L  = 1500 × 8 = 12,000 bits
R  = 10 Mbps  = 10,000,000 bps
Tt = 12,000 / 10,000,000 = 0.0012 s = 1.2 ms

Transmission delay depends only on packet size and link rate. It has nothing to do with how long the cable is.

2. Propagation Delay

Once a bit is on the wire, it still needs time to physically travel to the other end. Signals in copper and optical fibre travel at roughly 2 × 108 m/s (about two-thirds of the speed of light). Radio signals in free space travel at about 3 × 108 m/s.

Propagation delay  Tp = d / s

Example: A 2,000 km fibre link.

d  = 2000 km = 2,000,000 m
s  = 2 × 10^8 m/s
Tp = 2,000,000 / 200,000,000 = 0.01 s = 10 ms

Propagation delay depends only on distance and the medium. It does not depend on the packet size or the data rate.

3. Queuing Delay

A router may receive packets faster than it can send them out on an outgoing link. Extra packets wait in a buffer (a queue). The time spent waiting is the queuing delay.

Arriving packets → [ P5 | P4 | P3 | P2 ] → Outgoing link
                     waiting in buffer

Queuing delay is the hardest to predict because it changes with traffic. A useful measure is the traffic intensity:

Traffic intensity = (L × a) / R

L = packet size in bits
a = average packet arrival rate (packets per second)
R = link rate (bps)
  • If traffic intensity is close to 0, packets rarely wait, so queuing delay is small.

  • As it gets close to 1, queuing delay grows very quickly.

  • If it is greater than 1, packets arrive faster than they can leave. The queue keeps growing, and once the buffer is full, new packets are dropped. This is one main cause of packet loss.

Example: 500 packets/s arrive, each 12,000 bits, on a 10 Mbps link: intensity = (12,000 × 500) / 10,000,000 = 0.6. The link is busy 60% of the time.

4. Processing Delay

When a router receives a packet, it must read the header, check for bit errors and look up where to forward it. On modern hardware this usually takes microseconds or less, so many exam problems tell you to ignore it.


Transmission Delay vs Propagation Delay

Students mix these two up more than anything else in this topic. Here is a simple analogy.

Imagine 10 cars must pass through a toll booth and then drive 100 km to the next city.

  • The time for all 10 cars to get through the toll booth is like transmission delay. More cars (bigger packet) or a slower booth (lower rate) means more time.

  • The time for one car to drive 100 km is like propagation delay. It depends only on the distance and the speed of the car.

Feature

Transmission Delay

Propagation Delay

Formula

L / R

d / s

Depends on

Packet size, link rate

Distance, medium

Bigger packet?

Increases

No change

Faster link (higher Mbps)?

Decreases

No change

Longer cable?

No change

Increases

Upgrading from 100 Mbps to 1 Gbps reduces transmission delay, but it cannot make light travel faster. A Mumbai to New York round trip stays well over 100 ms no matter how much bandwidth you buy.


Real paths have several links connected by routers. Routers normally use store-and-forward (you will see this again in the Switching chapter): a router must receive the entire packet before it can start sending it on the next link.

Source ──link 1──→ Router A ──link 2──→ Router B ──link 3──→ Destination

If there are N identical links and we ignore queuing and processing:

Delay for one packet = N × (Tt + Tp)

Example: One 1,500-byte packet over 3 links, each 10 Mbps, and ignore propagation:

Delay = 3 × 1.2 ms = 3.6 ms

When P packets are sent back to back, they move through the links like a pipeline. The first packet takes N × Tt to arrive, and each next packet arrives Tt later. Ignoring propagation, queuing and processing:

Time for P packets over N links = (N + P − 1) × Tt

Example: 5 packets over 3 links, Tt = 1.2 ms each:

(3 + 5 − 1) × 1.2 ms = 7 × 1.2 ms = 8.4 ms

If propagation is not ignored, add the propagation delay of every link once: (N + P − 1) × Tt + N × Tp for identical links.


Latency, RTT, Throughput and Bandwidth

These words are used loosely in daily life, but in networking they have specific meanings.

Term

Meaning

Bandwidth (link rate)

The maximum rate a link can send bits, e.g. 100 Mbps

Throughput

The actual rate at which data is successfully delivered, which is often less than bandwidth

Latency

How long it takes data to get from sender to receiver (one-way delay)

RTT (Round-Trip Time)

Time for a message to go to the receiver and a reply to come back

If the delay is the same in both directions and the reply is tiny, RTT is roughly 2 × propagation delay plus the transmission, queuing and processing delays along the way. The ping command shows RTT.

When data passes through several links, the end-to-end throughput can never be higher than the slowest link on the path.

Server ──100 Mbps──→ Router ──20 Mbps──→ Your home

Throughput ≤ min(100, 20) = 20 Mbps

Example: Downloading a 50 MB file (here 1 MB = 106 bytes) over this path, ignoring delays and other users:

File size = 50 × 10^6 × 8 = 400 × 10^6 bits
Time      = 400 × 10^6 / (20 × 10^6) = 20 seconds

Bandwidth-Delay Product

The bandwidth-delay product tells you how many bits can be "in flight" on a link at the same time.

Bandwidth-delay product = R × Tp

Example: For our 10 Mbps, 2,000 km link with Tp = 10 ms:

10,000,000 × 0.01 = 100,000 bits are on the wire at once

Think of the link as a pipe: bandwidth is how wide it is and propagation delay is how long it is. The product is the volume of the pipe. Some books use R × RTT instead, which counts the bits a sender must send before the first acknowledgment can come back. That version becomes very important when we study sliding windows in the ARQ Protocols chapter: to keep a long, fast link busy, the sender needs a large window.


Channel Capacity

So far we treated the link rate R as a given number. But why can a telephone line carry only tens of kilobits per second, while a fibre carries gigabits? Two results answer this.

Bandwidth in Hertz vs Bits per Second

In this section, bandwidth B means the range of frequencies a channel can pass, measured in Hertz (Hz). For example, a classic voice telephone line passes roughly 300 to 3,300 Hz, so B ≈ 3,000 Hz. The capacity C is in bits per second.

Signal Levels, Baud Rate and Bit Rate

A transmitter sends symbols (signal elements). If each symbol can take one of M different levels, each symbol carries log2M bits.

Bit rate = Baud rate × log2(M)

Baud rate = symbols per second
M = number of signal levels

Example: 1,000 symbols per second with 16 levels gives 1,000 × log216 = 1,000 × 4 = 4,000 bps.

Nyquist Formula (Noiseless Channel)

For a perfect channel with no noise, Nyquist showed that a channel of bandwidth B can carry at most 2B symbols per second. So:

Maximum bit rate = 2 × B × log2(M)

Example: A noiseless channel with B = 3,000 Hz using 4 signal levels:

2 × 3000 × log2(4) = 2 × 3000 × 2 = 12,000 bps

The formula suggests we could get any rate by adding more levels. In real life, noise makes closely spaced levels hard to tell apart. That is where Shannon comes in.

Shannon Capacity (Noisy Channel)

Shannon's formula gives the theoretical maximum error-free data rate of a noisy channel:

C = B × log2(1 + SNR)

C   = capacity in bits per second
B   = bandwidth in Hz
SNR = signal-to-noise ratio (a plain ratio, NOT in dB)

SNR is often given in decibels (dB). Convert it first:

SNR_dB = 10 × log10(SNR)      →      SNR = 10^(SNR_dB / 10)

10 dB → 10      20 dB → 100      30 dB → 1000

Example: A telephone line with B = 3,000 Hz and SNR = 30 dB.

SNR = 10^(30/10) = 1000
C   = 3000 × log2(1 + 1000)
    = 3000 × log2(1001)
    ≈ 3000 × 9.967
    ≈ 29,902 bps  (about 30 kbps)

No matter how clever the encoding is, this line cannot reliably carry more than about 30 kbps. Real systems get close to, but never above, the Shannon limit.

Using Both Together

A classic exam question uses Shannon to find the upper limit and Nyquist to find how many signal levels are needed.

Example: A channel has B = 1 MHz and SNR = 63 (plain ratio).

Step 1 (Shannon): C = 10^6 × log2(1 + 63) = 10^6 × log2(64) = 6 Mbps

Step 2: For a safety margin, choose a lower rate, say 4 Mbps.

Step 3 (Nyquist): 4 × 10^6 = 2 × 10^6 × log2(M)
                  log2(M) = 2
                  M = 4 signal levels

Feature

Nyquist

Shannon

Channel type

Noiseless (ideal)

Noisy (real)

Formula

2B log2M

B log2(1 + SNR)

Depends on

Bandwidth and number of levels

Bandwidth and signal-to-noise ratio

Tells you

Max bit rate for a chosen number of levels

Absolute upper limit for any encoding


A Small Delay and Capacity Calculator

Here is a short Python program that computes all the examples from this chapter. Changing the numbers is a good way to check your own homework answers.

import math

def transmission_delay(bits, rate_bps):
    return bits / rate_bps                  # seconds

def propagation_delay(distance_m, speed_mps=2e8):
    return distance_m / speed_mps           # seconds

def shannon_capacity(bandwidth_hz, snr_db):
    snr = 10 ** (snr_db / 10)               # dB -> plain ratio
    return bandwidth_hz * math.log2(1 + snr)

def nyquist_bit_rate(bandwidth_hz, levels):
    return 2 * bandwidth_hz * math.log2(levels)

# Example 1: one 1500-byte packet on a 10 Mbps, 2000 km link
L = 1500 * 8                                # bits
R = 10_000_000                              # bits per second
tt = transmission_delay(L, R)
tp = propagation_delay(2000 * 1000)
print(f"Transmission delay : {tt * 1000:.2f} ms")
print(f"Propagation delay  : {tp * 1000:.2f} ms")
print(f"Total (no queuing) : {(tt + tp) * 1000:.2f} ms")
print(f"Bandwidth-delay    : {R * tp:,.0f} bits on the wire")

# Example 2: telephone line, B = 3000 Hz, SNR = 30 dB
print(f"Shannon capacity   : {shannon_capacity(3000, 30):,.0f} bps")

# Example 3: noiseless 3000 Hz channel with 4 signal levels
print(f"Nyquist bit rate   : {nyquist_bit_rate(3000, 4):,.0f} bps")

Output:

Transmission delay : 1.20 ms
Propagation delay  : 10.00 ms
Total (no queuing) : 11.20 ms
Bandwidth-delay    : 100,000 bits on the wire
Shannon capacity   : 29,902 bps
Nyquist bit rate   : 12,000 bps

Common Mistakes

  • Forgetting to convert bytes to bits. A 1,000-byte packet is 8,000 bits. Missing the × 8 makes your answer 8 times too small.

  • Using the SNR in dB directly in Shannon's formula. 30 dB means a ratio of 1,000, not 30.

  • Thinking propagation delay depends on bandwidth. A faster link sends bits out sooner, but each bit still travels at the same speed.

  • Confusing bandwidth in Hz with bandwidth in bps. Nyquist and Shannon use Hz; delay formulas use bps.

  • Adding transmission delay only once over many links. With store-and-forward, each link adds its own transmission delay.

  • Thinking Shannon capacity is the speed you will actually get. It is an upper limit; real throughput is lower.


Interview Tips

  • If asked "Why is my 1 Gbps connection still slow to a far-away server?", talk about propagation delay and RTT. Bandwidth helps with large downloads; latency matters for small requests, games and video calls.

  • Be able to explain the four delays in one sentence each, and say which ones are fixed (propagation, transmission for a given packet) and which vary (queuing).

  • Know that packet loss in the Internet mostly happens when router queues overflow, which links queuing delay to congestion.

  • Remember "Nyquist = noiseless, levels" and "Shannon = noisy, SNR".

Conclusion

Every packet faces four delays at each hop: processing, queuing, transmission (L/R) and propagation (d/s). Transmission delay depends on packet size and link rate, while propagation delay depends on distance and the medium.

End-to-end throughput is limited by the bottleneck link, and the bandwidth-delay product tells you how much data fits in the pipe.

The maximum data rate of a channel is given by Nyquist for a noiseless channel (2B log2M) and by Shannon for a noisy channel (B log2(1 + SNR)).

The easiest way to remember it is:

Transmission = how fast you push bits out. Propagation = how far they have to travel. Capacity = the most the channel can ever carry.