Bandwidth, Latency, and Jitter: Three Different Network Speeds

A connection can download a large file quickly and still feel poor during a game. A video call can break up even when a speed test reports hundreds of megabits per second.

The word "speed" hides several independent dimensions.

Bandwidth describes capacity, latency describes delay, and jitter describes variation in delay.

Packet loss, congestion, and application behavior interact with all three.

Bandwidth

Bandwidth is the maximum data-carrying capacity of a link or path, usually measured in bits per second.

Examples:

100 Mbps
1 Gbps
10 Gbps

The term is sometimes used loosely for observed transfer rate, but throughput is the actual achieved rate.

A 1 Gbps Ethernet link does not guarantee a 1 Gbps internet download. Every stage in the path has its own capacity.

The bottleneck limits end-to-end throughput.

Latency

Latency is the delay from an action or packet to a result.

One-way latency measures one direction. Round-trip time, or RTT, measures a trip to a destination and back.

Sources include:

• Physical propagation distance
• Serialization onto links
• Router and switch processing
• Queue waiting
• Radio scheduling and retries
• Server processing

Light in fiber travels slower than light in vacuum. A long-distance path has an unavoidable physical delay even with enormous bandwidth.

Jitter

Jitter is variation in packet delay.

Suppose voice packets arrive after:

30 ms, 31 ms, 29 ms, 95 ms, 32 ms

Average latency may look acceptable, but the 95 ms packet disrupts smooth playback.

Real-time applications use a jitter buffer, waiting briefly so packets with varying arrival times can be played in order.

A larger buffer conceals more variation but adds conversational delay.

Packet loss

Packets can be dropped because of:

• Congestion
• Wireless interference
• Faulty links
• Buffer limits
• Routing problems

Reliable protocols retransmit, adding delay and reducing throughput.

Real-time media may not wait for an old packet. It can conceal loss using interpolation, repeat prior audio, or reduce quality.

Even a small loss rate can heavily affect high-throughput TCP over long-latency paths.

A concrete comparison

Consider two connections.

Connection A

500 Mbps bandwidth
100 ms RTT
stable delay

Connection B

100 Mbps bandwidth
10 ms RTT
stable delay

Connection A can download one large file faster once data flows.

Connection B can feel more responsive for:

• Game controls
• Remote desktop
• Many small web requests
• Interactive command-line work

Neither is universally "faster."

Serialization delay

Packets take time to place onto a link.

For a 1500-byte packet:

1500 bytes × 8 = 12,000 bits

On a 1 Mbps link:

12,000 ÷ 1,000,000 = 12 ms

On a 1 Gbps link:

12,000 ÷ 1,000,000,000 = 0.012 ms

Slow links add noticeable per-packet serialization time.

Queuing delay

When packets arrive faster than a link can send them, they queue.

During a large upload, a router can build a long queue. Small game or voice packets wait behind bulk data.

This is bufferbloat: excessive buffering produces high latency under load.

Smart queue management controls the sending rate just below the bottleneck and fairly schedules flows, often improving responsiveness dramatically.

A speed test should measure latency while idle and under upload/download load.

Why web pages care about latency

A page may require:

• DNS lookup
• Connection handshake
• Encryption handshake
• HTML request
• Additional resource requests

Modern protocols reduce round trips and multiplex requests, but latency still accumulates.

A page with many dependencies can feel slow even when its total byte size is modest.

Content-delivery networks place data geographically and topologically closer to users partly to reduce latency.

Bandwidth-delay product

A high-bandwidth, high-latency path needs substantial data in flight to use its capacity.

The amount is approximately:

bandwidth × round-trip time

For 1 Gbps and 100 ms:

1,000,000,000 bits/s × 0.1 s
= 100,000,000 bits
= 12.5 MB

The transport protocol may need at least that much unacknowledged data in flight to fill the path.

Window sizes and congestion control influence whether a single connection reaches maximum throughput.

Measuring performance

Useful tools and metrics include:

• Ping for RTT and loss indications
• Traceroute for path visibility
• Speed tests for throughput
• Continuous probes for jitter
• Application metrics for actual experience

Ping traffic can be filtered or handled differently from application traffic.

A nearby speed-test server may not represent a distant service.

Measure from the affected device, during the affected workload, to relevant destinations.

Wired and wireless variation

Ethernet on a healthy local link tends to offer stable latency and low loss.

Wi-Fi and cellular share radio spectrum, so scheduling, interference, movement, and retries create more variation.

One momentary radio disturbance may not change average bandwidth much but can produce jitter that affects a call.

Applications often adapt bitrate or buffer length based on changing conditions.

Common misunderstandings

"High Mbps means low ping"

Capacity and latency are separate. A high-bandwidth path can still have long propagation or queuing delay.

"Average latency tells the whole story"

Jitter and high-percentile delays reveal intermittent stalls hidden by the average.

"More buffering always prevents loss and improves performance"

Large buffers can reduce drops but create severe delay under load.

"A speed test proves every service should be fast"

It tests one path, server, protocol, and moment. Other services can have different routes or bottlenecks.

Knowledge check

The one idea to remember

Network quality is multidimensional.

Bandwidth controls capacity, latency controls response delay, jitter controls consistency, and loss triggers recovery or visible degradation. Evaluate the dimensions your application actually needs.

Next, we will see how transport ports direct packets from a host address to the correct application conversation.