Networking — Trivia & Interesting Facts¶

Surprising, historical, and little-known facts about computer networking.

The first ARPANET message crashed the system after two letters¶

On October 29, 1969, Charley Kline at UCLA attempted to send the word "LOGIN" to Stanford Research Institute over the ARPANET. The system crashed after transmitting only "LO" — making "Lo" the first message sent over the precursor to the Internet. The full "LOGIN" was successfully sent about an hour later after the system was restarted.

The OSI model was supposed to win, and TCP/IP was considered the underdog¶

In the 1980s, the International Organization for Standardization (ISO) developed the OSI protocol suite as the "proper" way to do networking, backed by governments and telecom companies worldwide. TCP/IP was viewed as an academic experiment. By the early 1990s, TCP/IP had won decisively because it was simpler, already worked on the ARPANET, and was freely available in BSD Unix. The OSI model survives only as a teaching framework — none of its actual protocols are used.

The 10.0.0.0/8 private address space contains more IPs than most countries have public addresses¶

RFC 1918 (1996) reserved three private address blocks: 10.0.0.0/8 (16.7 million addresses), 172.16.0.0/12 (1 million addresses), and 192.168.0.0/16 (65,536 addresses). The 10.0.0.0/8 space alone is larger than the public IPv4 allocations of most countries. These private ranges, combined with NAT, are how the Internet scaled far beyond its original 4.3 billion address limit.

The speed of light limits network latency to about 3.3 microseconds per kilometer in fiber¶

Light travels at approximately 200,000 km/s in fiber optic cable (about 2/3 the speed of light in vacuum due to the refractive index of glass). This means the minimum one-way latency from New York to London (5,500 km of submarine cable) is about 27 milliseconds, and no amount of engineering can reduce it. This physics constraint is why high-frequency trading firms spend billions on shorter cable routes — saving even 1 millisecond of latency is worth millions in revenue.

The Internet nearly ran out of BGP route table memory in 2014¶

On August 12, 2014, the global BGP routing table exceeded 512,000 prefixes — the maximum that older Cisco routers with 512 MB TCAM could hold. Routers at Verizon, AT&T, and other major ISPs crashed or began dropping routes, causing widespread Internet disruptions. The trigger was a Verizon customer that accidentally leaked approximately 15,000 new prefixes. This event, sometimes called "512K Day," was the networking equivalent of Y2K actually happening.

Ethernet was invented at Xerox PARC and named after "luminiferous ether"¶

Robert Metcalfe invented Ethernet at Xerox PARC in 1973, naming it after the "ether" — the hypothetical medium that 19th-century physicists believed carried light waves. The original Ethernet ran at 2.94 Mbps over thick coaxial cable. Metcalfe went on to found 3Com and later predicted (incorrectly) that the Internet would collapse in 1996. He ate his words — literally, by blending a printed copy of his column into a smoothie and drinking it on stage.

TCP slow start isn't slow — it's exponential¶

Despite its name, TCP slow start doubles the congestion window with every round trip, producing exponential growth. Starting from 1 segment, after 10 RTTs you're sending 1,024 segments. The name "slow start" comes from the fact that it's slower than what TCP did before — which was to blast the full window immediately, causing network-wide congestion collapses. Van Jacobson's 1988 congestion control algorithms (slow start and congestion avoidance) literally saved the early Internet from collapsing.

The maximum theoretical throughput of a network is limited by bandwidth-delay product¶

The bandwidth-delay product (BDP) determines how much data can be "in flight" between two endpoints. For a 10 Gbps link with 100ms RTT (e.g., cross-country), BDP = 125 MB. The TCP receive window must be at least this large to keep the pipe full, which is why TCP window scaling (RFC 1323) was essential — the original 16-bit window field maxed out at 64 KB, limiting throughput to about 5 Mbps on high-latency paths regardless of link speed.

Submarine cables carry 99% of intercontinental data traffic¶

Despite the rise of satellite Internet (Starlink, etc.), over 99% of intercontinental data travels through approximately 550 submarine fiber optic cables on the ocean floor. These cables are about the diameter of a garden hose, are regularly damaged by ship anchors and fishing trawlers, and are repaired by specialized cable ships. The total length of submarine cables exceeds 1.4 million kilometers — enough to circle the Earth 35 times.

The "five nines" of availability (99.999%) allows only 5.26 minutes of downtime per year¶

Network SLAs are often expressed in "nines." 99.9% (three nines) allows 8.76 hours of annual downtime. 99.99% (four nines) allows 52.6 minutes. 99.999% (five nines) allows only 5 minutes and 15 seconds. Achieving five nines for a network path requires full redundancy at every layer — dual power, dual switches, dual uplinks, dual providers, and automated failover faster than any human can react.