Circuit Switching vs Packet Switching

Section 01

The Story That Explains Everything

📖 Real World Analogy

The Grand Trunk Road vs the Indian Postal Service

Imagine you need to send an urgent message from Amritsar to Kolkata — roughly 1,900 km away. You have two very different options.

Option A — Reserve a Private Highway. You call the government and demand that a dedicated lane on the entire Grand Trunk Road be reserved only for you, from Amritsar all the way to Kolkata. Nobody else may drive on it while your message is en route. Once your courier finishes the trip, the lane is released. Result — smooth, uninterrupted, but massively wasteful. The lane sits empty most of the time.

Option B — Break the message into 100 postcards. You cut your letter into 100 numbered postcards and drop each one in the postbox. Every postcard finds its own path — some go by train, some by truck, some by air. They arrive out of order. At the receiving end, your friend reassembles them by their numbers. Roads stay shared. Everybody benefits.

Option A is Circuit Switching. Option B is Packet Switching. Every message you send on the modern internet uses Option B.

These two ideas — born decades apart — define how every voice call, video stream, WhatsApp message and web page moves through the world. Understanding them (and the four metrics that measure their quality) is the foundation of every conversation about computer networks.

🌐

The Core Insight

Circuit switching reserves a path before you send anything. Packet switching chops the message into pieces and lets each piece find its own way. Reservation gives you predictability at the cost of efficiency. Chopping gives you efficiency at the cost of predictability. The internet chose efficiency — and then spent 50 years inventing techniques to buy back the predictability.

Section 02

Circuit Switching — The Old Telephone Way

Circuit switching is the original way networks moved information. When you picked up a landline telephone in 1975 and dialled a number, the phone exchange physically closed a series of switches to form one continuous electrical path from your handset to the person you were calling. That path stayed reserved for you until one of you hung up.

📞 The Three Phases of a Circuit-Switched Call

Phase 1

Circuit Setup — The network searches for a free path from A to B and reserves every switch and wire along it. This takes time (a few hundred milliseconds).

Phase 2

Data Transfer — Your voice flows continuously along the reserved circuit. Delay is minimal and constant. Quality is guaranteed by physics.

Phase 3

Circuit Teardown — When you hang up, the network releases every switch. Someone else can now use that path.

⚠️

Why Circuit Switching Is Wasteful

In a normal phone conversation, roughly 60–70% of the time is silence — you pausing between sentences, thinking, or listening. During every second of that silence, the entire reserved circuit sits idle, forbidden to anybody else. On a wire that could carry hundreds of simultaneous calls, this is enormously wasteful. Circuit switching is predictable, but it pays for that predictability in unused capacity.

Section 03

Packet Switching — The Internet Way

Packet switching was invented in the 1960s by Paul Baran and Donald Davies as a way to build a communication network that could survive a nuclear war. Instead of reserving paths, it does the opposite: it chops every message into small, independent pieces called packets, and dumps them into a shared network to find their own way.

📦 Anatomy of a Packet

Header

Source address, destination address, sequence number, checksum — the “envelope” information.

Payload

A small chunk of the actual data (typically 1,000–1,500 bytes). One video frame might be split into dozens of these.

Trailer

Error-detection bits so the receiver can verify the packet arrived intact.

Each packet is stamped with the destination, dropped into the network, and left to find its own route. Routers along the way inspect each packet's destination address and forward it to the next hop — a process called store-and-forward. Two packets from the same message can (and often do) take totally different routes and arrive in the wrong order. The receiver reassembles them using their sequence numbers.

🎯

Why This Won the World

Packet switching uses network capacity only when there is something to send. No idle silence wastes anybody's bandwidth. A single fibre-optic cable can carry millions of simultaneous conversations, video streams, and downloads because each is just a stream of packets sharing the same wire. This is why the internet scales and old telephone networks did not.

Section 04

Animated Diagram — Watch Them Move

The animation below shows the same message travelling both ways. In the top track, a single reserved circuit carries one continuous stream from sender to receiver. In the bottom track, four packets take independent routes through a mesh of routers — sometimes overtaking each other, sometimes arriving out of order.

Circuit vs Packet — Live Animation

Top — a single dedicated circuit; every data unit flows through the same lane. Bottom — four packets belonging to the same message, each taking a different route through the router mesh. Notice how they can overtake each other.

Section 05

Side-by-Side Comparison

📞 Circuit Switching

Property	Behaviour
Path	Reserved end-to-end before sending
Setup time	Slow (100–500 ms)
Delay after setup	Constant, very small
Order of arrival	Always in order
Wastes capacity?	Yes, when idle
Handles failure	Poorly — call drops
Used by	Old landline phones, ISDN

📧 Packet Switching

Property	Behaviour
Path	Chosen per packet, dynamically
Setup time	None — just send
Delay after setup	Variable (jitter possible)
Order of arrival	May be out of order
Wastes capacity?	No — shared efficiently
Handles failure	Excellent — reroutes automatically
Used by	The entire modern internet

🔑

The Hybrid You Actually Use

Modern voice calls on WhatsApp, Google Meet or a mobile 5G network are not circuit-switched. They are packet-switched, but layered with clever techniques (jitter buffers, forward-error-correction, priority queues) to give them the smooth, reliable feel of a circuit. The best of both worlds — efficiency of packets, quality of circuits.

Section 06

Real Case — When Packet Switching Saves You

📰 News Case — Red Sea Cable Cuts, 2024–2025

How the Internet Rerouted Itself in Minutes

In 2024 and again in September 2025, multiple submarine internet cables in the Red Sea — carrying data between Europe, Africa and Asia — were cut. Cloudflare's Q3 2025 internet disruption report noted that a single cable set can carry a huge fraction of a country's international traffic. When a cable snapped near Jeddah, Saudi Arabia, on 6 September 2025, partial bandwidth was lost within seconds.

Yet the internet did not go down. Why? Because packet switching allowed routers across the world to reroute traffic through alternate paths — via the Suez, or all the way around Africa — within minutes. Users saw higher latency (their packets were now taking longer routes) but connectivity survived.

A circuit-switched network in the same situation would have dropped every live call the instant the cable failed — because the reserved circuits inside that cable no longer existed. Packet switching is the reason the modern internet is far more resilient than the phone network ever was.

Section 07

Now — How Do We Measure Network Quality?

A packet-switched network can carry your data brilliantly or terribly, and everything in between. To decide which, engineers use four fundamental metrics. Every single network problem you have ever experienced — a laggy video call, a stuttering IPL stream, a slow Google search — is a story told by these four numbers.

🔌

Bandwidth

Capacity of the pipe

The maximum data that could flow per second, measured in bits per second (bps, Mbps, Gbps). It's the width of the road, not the speed of cars on it.

⏱️

Latency

Delay per trip

The time (ms) taken for one packet to travel from source to destination. Determined by distance, medium (fibre vs satellite), and number of router hops.

🔊

Throughput

Actual data delivered

The real data rate you achieve after packet loss, retransmissions, and congestion. Always ≤ bandwidth — often much less.

🍀

Jitter

Variation in latency

The inconsistency of packet arrival times. A stream with 40 ms average latency but wild swings from 5 to 200 ms has awful jitter — and audio will crackle.

Section 08

Metric 1 — Bandwidth

📖 Story

The Water Pipe

Think of bandwidth as the diameter of a water pipe. A garden hose can push a certain maximum volume of water per minute; a fire hydrant can push massively more. Bandwidth is that maximum capacity — not the water currently flowing, but the most that could flow if the tap were fully open. Whether you actually use it all is a separate question.

Bandwidth is measured in bits per second and its multiples: Kbps, Mbps, Gbps, Tbps. When your ISP sells you a “100 Mbps fibre plan”, they are selling you the theoretical maximum bit rate on the link between your home and their nearest exchange. It says nothing about the speed you will actually experience while watching Netflix.

Bandwidth Tier	Typical Use	Real-World Example
1–5 Mbps	Web browsing, WhatsApp calls, SD video	Basic Jio or Airtel 4G in a weak-signal area
25–50 Mbps	HD video, casual gaming, one work-from-home user	Entry-level home broadband in India
100–300 Mbps	4K streaming, multiple users, video calls + downloads	Jio Fiber & Airtel Xstream Fiber typical plan
1 Gbps+	Whole household, 4K on multiple screens, cloud backups	Premium fibre and 5G Fixed Wireless
400 Gbps	Backbone links between data-centre regions	Undersea cable segments, Tier-1 ISP peering

⚠️

Advertised Bandwidth ≠ Delivered Bandwidth

Your ISP's “up to 100 Mbps” is a ceiling, not a promise. Wi-Fi loss inside your house, evening peak congestion at the ISP, undersized backhaul to the wider internet, and the server on the other end all conspire to reduce what you actually see. Bandwidth is a necessary but not sufficient ingredient of a fast connection.

Section 09

Metric 2 — Latency

📖 Story

The London Cab vs the Delhi Metro

You and a friend both need to cross town. You take a cab from Punjabi Bagh to Connaught Place; your friend takes the metro. The cab's capacity is 4 passengers; the metro carries thousands. Yet in Delhi rush hour, the metro often arrives first — because the cab is stuck in traffic (latency).

Bandwidth is how many passengers per hour. Latency is how long one particular passenger waits from start to finish. They are completely separate. A satellite link can have huge bandwidth and terrible latency, because the signal has to travel to space and back.

Latency has four components. Understanding them is the difference between guessing and actually diagnosing a slow network.

⏱️ The Four Sources of Latency

Propagation

Time for signal to physically travel. Speed of light in fibre is ~200,000 km/s. Amritsar to New York is ~12,000 km ⇒ ~60 ms one way, unavoidable by physics.

Transmission

Time to push all the bits onto the wire. Sending a 1,500-byte packet on a 10 Mbps link takes 1.2 ms. On a 1 Gbps link it's 12 μs. Higher bandwidth reduces this component.

Processing

Time each router takes to inspect the packet header and decide the next hop. Typically microseconds per hop, but adds up over 10–20 hops.

Queuing

Time the packet waits in a router's buffer because other packets are being sent ahead of it. This is the unpredictable component and the main cause of jitter.

📈

Practical Latency Numbers You Can Feel

Under 30 ms — competitive online gaming feels instant. 30–100 ms — video calls feel natural. 100–300 ms — noticeable delay, awkward interruptions in conversation. Over 500 ms — voice calls become almost unusable (satellite internet like older Hughesnet, ~600 ms). Starlink dropped this to ~40 ms by flying satellites much lower.

Section 10

Metric 3 — Throughput

Bandwidth is the theoretical maximum. Throughput is what you actually get. It's the honest, real-world data rate a connection sustains, measured over some meaningful interval. Throughput is always less than or equal to bandwidth — and in practice, often much less.

📡

Congestion

Other users on shared wire

A Jio 5G tower serving 500 phones divides its capacity among them. Evening 8–11 pm throughput drops sharply as neighbours join Netflix and Instagram Reels.

❌

Packet Loss

Dropped → retransmit

Every lost packet must be resent, doubling its cost. TCP also slows down transmission when it detects loss (congestion control). 1% loss can halve throughput.

🛠️

Protocol Overhead

Headers & handshakes

Every packet carries TCP/IP headers (~40 bytes) plus TLS encryption. On short flows, handshake round-trips can dominate. Real throughput is often 70–90% of link capacity.

Bandwidth vs Throughput — The Gap

Same 100 Mbps connection, three different throughput values across a single evening. The pipe's capacity does not change — what flows through it does.

Section 11

Metric 4 — Jitter

📖 Story

The Drummer Who Cannot Keep Time

Imagine a drummer who is supposed to hit the drum every 20 milliseconds. On average, they do — but individual beats come at 5 ms, then 40 ms, then 8 ms, then 35 ms. The average is correct. The rhythm is destroyed. Every listener notices immediately.

That's jitter — not the absolute delay of packets, but the variation from packet to packet. For streaming voice and video, jitter is often worse than latency. A steady 150 ms delay is easy to work with; a wildly varying 50–200 ms delay makes audio warble and video freeze.

Latency Pattern — Smooth vs Jittery

Both connections have the same average latency (~120 ms). The green connection is usable for video calls; the red one will cause constant robotic-sounding audio and video freezes.

🔑

The Jitter Buffer — How Apps Fight Back

WhatsApp, Google Meet and Zoom all use a jitter buffer: they hold incoming audio packets for a short time (typically 50–200 ms) before playing them, so that late-arriving packets have time to catch up. Bigger buffer = smoother audio but more delay. Smaller buffer = lower delay but more crackles. It's a live trade-off, adjusted every second as your network changes.

Section 12

All Four Metrics on One Timeline

Bandwidth — the pipe capacity

Set by the physical link — fibre, copper, radio. The one number your ISP quotes on the bill. Static, changes only when you upgrade your plan or install new hardware.

Latency — how long one trip takes

Determined mostly by distance and hop count. Cannot be reduced below the speed of light. Your ping in Valorant is dominated by this.

Throughput — how much actually gets through

A function of bandwidth, latency, packet loss, and protocol overhead. What speedtest.net actually measures. Changes second by second.

Jitter — how steady the arrivals are

The standard deviation of latency. Invisible until it isn't. The single biggest factor in real-time application quality.

Section 13

Real Case — The 2025 AWS Outage

📰 News Case — October 2025

One DNS Failure Broke 3,500 Companies in 60 Countries

On 20 October 2025, Amazon Web Services' US-EAST-1 region experienced a major outage traced to a DNS resolution failure. In just two hours, over 4 million outage reports were submitted globally — peak reports hit 6.3 million from the US and 1.5 million from the UK. Snapchat alone received ~3 million outage reports; Downdetector recorded a 970% surge in daily baseline reports during the peak.

What did users actually experience? The bandwidth of their home connection didn't change. Their fibre still promised 300 Mbps. But throughput collapsed to nearly zero for anything hosted on AWS — because packets sent to those servers simply timed out. Latency to Amazon-hosted services became infinite (no response). Jitter became irrelevant — when no packets arrive, there is nothing to measure the variance of.

The lesson: end-user metrics depend on the health of every hop in the path, not just your local link. A single misconfigured DNS record on the other side of the planet can zero out your real-world throughput.

📰 News Case — Microsoft, July 2024

When Retry Storms Cause Their Own Congestion

In July 2024, Microsoft 365 users saw a widespread outage. ThousandEyes observed increased packet loss at the edge of the Microsoft network and heavy congestion. Microsoft's post-mortem said the trigger was a configuration change that caused an “influx of retry requests”. Each failed request produced automatic retries; the retries filled router queues; the queues caused delays; the delays triggered more retries. This is a classic congestion cascade — a live demonstration of what happens when a packet-switched network's shared queues become the bottleneck. Every metric — latency, throughput, jitter — degraded together.

Section 14

Practical Example — Diagnosing a Bad Video Call

Your Google Meet call keeps freezing. You have a 300 Mbps fibre plan. How do you actually figure out what's wrong? Walk through the four metrics in order and each one narrows the possibilities.

🕵️ The Four-Metric Diagnostic

Check 1

Run a speed test — are you getting close to advertised bandwidth? If you paid for 300 Mbps and Speedtest shows 12 Mbps, the problem is upstream (ISP congestion, Wi-Fi, or a bad cable). Video calls only need ~3 Mbps, so this alone rarely explains freezes.

Check 2

Ping the Google Meet server — measure latency. Under 100 ms is fine. Over 250 ms means every voice packet is arriving too late to sound natural. This usually points to distance or an overloaded router near you.

Check 3

Run a sustained transfer — check for packet loss. Loss over 1% will cause visible freezes because TCP retransmissions and forward-error-correction can't keep up. Common culprit: a weak Wi-Fi signal (move closer to the router).

Check 4

Watch ping variation over 30 seconds — jitter. If ping wildly swings between 20 ms and 400 ms, the problem is queueing somewhere. Usually another device on your Wi-Fi is saturating the uplink (a phone uploading photos, an OS update downloading in the background).

EXAMPLE — running these checks on a real slow Meet call

Speedtest: 284 Mbps down / 31 Mbps up <- bandwidth fine Ping to meet.google.com: min = 24 ms avg = 187 ms max = 620 ms loss = 4% ^^^^^ ^^^ huge jitter loss too high Diagnosis: Someone in the house is saturating the uplink. Culprit: Phone auto-backup uploading 3 GB of photos to iCloud. Fix: Pause the backup. Latency drops to 28 ms, loss to 0%.

💡

The Video-Call Rule of Thumb

HD video calls need only ~3 Mbps of bandwidth — but they demand <150 ms latency, <30 ms jitter, and <1% loss. Video is far more sensitive to the last three than to the first. Upgrading from a 100 Mbps to a 1 Gbps plan won't fix a jittery call. Fixing the Wi-Fi will.

Section 15

Which Metric Matters for Which Application?

Application	Bandwidth	Latency	Throughput	Jitter
Downloading a 20 GB game	Critical	Minor	Critical	Minor
Watching a 4K Netflix film	Important	Minor	Important	Buffered away
Google Meet video call	Modest need	Critical	Important	Critical
Online competitive gaming (BGMI, Valorant)	Modest need	Critical	Modest	Critical
Live IPL streaming	Important	Important	Important	Important
Browsing news websites	Modest	Important	Modest	Minor
Stock trading terminal	Modest	Critical	Modest	Critical
WhatsApp voice call	Tiny	Critical	Modest	Critical

🏆

The Big Idea in One Sentence

Batch tasks care about bandwidth. Real-time tasks care about latency and jitter. When you're deciding whether to upgrade a network or debug a slow app, first ask which of these two families the task belongs to — and the right answer usually becomes obvious.

Section 16

Golden Rules

🔑 Non-Negotiable Rules for Network Understanding

Packet switching won. Circuit switching still exists (some legacy telecom backbones, ISDN in industry) but every consumer-facing network on Earth now uses packets. Understand it as your default.

Bandwidth is the pipe. Throughput is the water. They are not the same thing. When someone says “the internet is slow,” they almost always mean throughput — not bandwidth — and the fix is usually not upgrading the plan.

Latency below ~30 ms is imperceptible; above ~250 ms is painful for interaction. Between those, applications differ — but the human brain has hard-wired expectations that no engineering trick can fully overcome.

You cannot beat the speed of light. Ping from Chandigarh to Sydney is ~150 ms and always will be. If a design requires 20 ms round-trip globally, redesign it or accept regional deployment.

Jitter is the assassin. Users rarely complain about steady latency, but they always notice when things stutter. If real-time app quality feels bad and average metrics look fine, always check jitter next.

Every network metric is measured end-to-end. Your throughput is limited by the slowest link in the chain — whether that's your Wi-Fi, your ISP's backhaul, an undersea cable, or the destination server's overloaded CPU. Diagnose the chain, not just your endpoint.

Congestion is invisible until it isn't. Router queues silently absorb load — and then, past a threshold, drop packets sharply. This is why performance often falls off a cliff rather than degrading smoothly. Design and monitor with headroom.