4.1 Types of Delay in Packet Networks

Types of Delay in Packet Networks

Packet networks experience four main types of delay: processing, queuing, transmission, and propagation. These delays occur at each hop as packets travel from source to destination, and together they determine the end-to-end latency a packet experiences. Understanding what causes each delay, how to calculate it, and what you can do to reduce it is central to network design and performance analysis.

Types of Delay in Packet Networks

Types of packet-switched network delays

Every router or switch along a packet's path introduces four distinct delays. They always occur in the same order at each hop: the device processes the packet, the packet waits in a queue, the device transmits the packet onto the outgoing link, and the signal propagates across that link.

Processing delay is the time a network device (router, switch) spends examining a packet before forwarding it. This includes checking for bit-level errors, looking up the destination in the forwarding table, and updating header fields like TTL and the header checksum. Processing delay depends on the device's CPU speed and how busy it is. On modern routers, processing delay is typically on the order of microseconds.

Queuing delay is the time a packet sits in a buffer waiting for its turn to be transmitted on the outgoing link. It occurs whenever packets arrive faster than the link can send them out. Of the four delays, queuing delay is the most variable. It depends on traffic load, link capacity, burstiness of arrivals, and the queuing discipline in use (FIFO, priority queuing, weighted fair queuing). Under light load, queuing delay can be nearly zero; under heavy load, it can dominate total latency.

Transmission delay is the time required to push all bits of a packet onto the link. It's purely a function of packet size and link bandwidth:

$d_{trans} = \frac{L}{R}$

where $L$ is the packet length in bits and $R$ is the link bandwidth in bits per second.

For example, transmitting a 1500-byte packet over a 100 Mbps link:

$d_{trans} = \frac{1500 \times 8}{100 \times 10^6} = 120 \; \mu s$

A common mistake is confusing transmission delay with propagation delay. Transmission delay is about pushing bits onto the wire; propagation delay is about bits traveling across the wire.

Propagation delay is the time for a single bit to travel from one end of a link to the other. It depends on the physical length of the link and the propagation speed of the medium:

$d_{prop} = \frac{d}{s}$

where $d$ is the link length in meters and $s$ is the propagation speed in meters per second.

For example, a 1000 km fiber optic link with a propagation speed of $2 \times 10^8$ m/s:

$d_{prop} = \frac{1000 \times 10^3}{2 \times 10^8} = 5 \; ms$

Notice that propagation delay doesn't depend on packet size at all, and transmission delay doesn't depend on link length. Keeping these two straight is one of the most important distinctions in this unit.

Calculation of end-to-end packet delay

To find the total delay a packet experiences from source to destination, you sum all four delay components at every hop along the path:

$d_{end\text{-}to\text{-}end} = \sum_{i=1}^{n} \left( d_{proc_i} + d_{queue_i} + d_{trans_i} + d_{prop_i} \right)$

where $n$ is the number of hops (links) on the path.

Step-by-step approach for a typical problem:

1. Identify the number of hops (routers/links) between source and destination.
2. For each hop, calculate or look up the four delay values.
3. Sum the four delays at each hop to get the per-hop delay.
4. Sum the per-hop delays across all hops to get the end-to-end delay.

For a simplified case where every hop has identical characteristics, this reduces to:

$d_{end\text{-}to\text{-}end} = n \times (d_{proc} + d_{queue} + d_{trans} + d_{prop})$

In practice, queuing delay varies from hop to hop and even from packet to packet, so real-world calculations often treat it statistically or use average values.

Contributions to network latency

Each delay type has a different relative impact depending on the network scenario:

• Processing delay is usually the smallest component on modern hardware (microseconds). It becomes significant only when devices are underpowered or performing complex tasks like deep packet inspection or encryption.
• Queuing delay is the most unpredictable. Under congestion, it can spike dramatically and become the dominant source of latency and jitter.
• Transmission delay matters most on low-bandwidth links or with large packets. On a high-speed backbone (10 Gbps+), transmission delay for a single packet is negligible. On a slow access link (e.g., 1 Mbps), it can be substantial.
• Propagation delay matters most on long-distance links. For a transcontinental or undersea fiber link, propagation delay alone can be tens of milliseconds, and there's no way to reduce it below the speed-of-light limit.

Factors influencing network delays

Processing delay factors:

1. Processing power of the device (CPU speed, available memory)
2. Complexity of per-packet operations (encryption, compression, deep inspection)

Strategies to reduce it:

• Use high-performance forwarding hardware (e.g., TCAM-based lookups)
• Offload complex tasks to dedicated hardware or distribute load across devices

Queuing delay factors:

1. Traffic load and burstiness of arrivals
2. Outgoing link capacity and current utilization
3. Queuing discipline (FIFO, priority queuing, weighted fair queuing)

Strategies to reduce it:

• Increase link capacity so utilization stays well below 100%
• Apply traffic shaping and policing to smooth out bursts
• Use priority queuing to protect delay-sensitive traffic like VoIP or video

Transmission delay factors:

1. Packet size (smaller packets transmit faster)
2. Link bandwidth (higher bandwidth means shorter transmission time)

Strategies to reduce it:

• Use higher-bandwidth links
• Use smaller packet sizes for delay-sensitive applications (though this increases header overhead)

Propagation delay factors:

1. Physical length of the link
2. Propagation speed of the medium (fiber ≈ $2 \times 10^8$ m/s; copper ≈ $2.3 \times 10^8$ m/s; these are close enough that medium choice matters less than distance)

Strategies to reduce it:

• Shorten the physical path (use more direct routing or place servers closer to users via CDNs)
• Choose media with higher propagation speed where possible, though the gains are modest compared to reducing distance