Why the access network is moving up and out
Everything in this ladder so far has assumed your first hop is a wire or a nearby router. But the hardest unsolved problem in networking is not the core — it is the last mile, the link that actually reaches a person. Half the planet has no usable fixed broadband, and stringing fiber to a mountain village or a fishing boat is hopeless. So the frontier of the access network is going two new places at once: up into low orbit, and out into denser, smarter cellular. Both are wireless, so both inherit every honest limit you met in the wireless rung — a radio still cannot detect a collision while it transmits, and a moving receiver still hands off between transmitters.
Keep the three quantities straight, because the marketing will try to blur them. Bandwidth is how wide the pipe is (bits per second, in principle). Throughput is how fast bits actually arrive once loss, sharing, and protocol overhead take their cut. Latency is the one-way or round-trip delay. They are different things, and here is the rule the sky cannot break: more bandwidth never reduces latency. A wider pipe lets you pour more water per second, but the first drop still has to travel the whole length at a finite speed. You cannot beat the speed of light, and as we are about to see, that single fact decides who wins the satellite race.
LEO: why the satellites came down low
Old satellite internet used geostationary satellites parked about 35,786 km up, hovering over one spot on the equator. The convenience is that your dish never has to move. The killer is distance: light takes roughly 120 ms just to climb that far, so a round trip up-and-back is about 240 ms before any processing — and a request-then-reply (your ping, the server's reply) crosses that gap twice, landing near half a second. That much propagation delay makes a video call painful and confuses TCP congestion control, whose careful speed-up depends on a short, steady round-trip time.
A low-earth-orbit (LEO) satellite network makes the opposite bet. Fly the satellites low — roughly 550 km — and the propagation delay collapses by a factor of sixty, putting round-trip times in the tens of milliseconds, competitive with ground links. The catch is geometry: a satellite that low is only overhead for a few minutes, and from 550 km it can only see a small patch of Earth. To cover the whole planet continuously you therefore need not one satellite but a constellation of thousands, with a handoff from one to the next every few minutes, exactly like a phone moving between cell towers — just at 27,000 km per hour.
Round trip = signal goes UP and comes back DOWN (one full there-and-back) light speed ~= 300,000 km per second GEO altitude ~35,786 km -> ~2 x 35,786 / 300,000 ~= 0.24 s (~240 ms) LEO altitude ~550 km -> ~2 x 550 / 300,000 ~= 0.004 s (~4 ms link) (real LEO RTT lands in the tens of ms once you add ground hops + processing)
5G is more than fast — it is sliceable
Down on the ground, the cellular network you met in the wireless rung keeps reinventing itself. The headline of 5G is speed, but the deeper story is flexibility. A 5G tower can use very high-frequency "millimeter wave" bands that carry enormous bandwidth over short, easily-blocked distances, alongside lower bands that travel far through walls — so coverage is now a deliberate trade rather than one fixed setting. Crucially, much of 5G's machinery moved from custom radio hardware into software running on general servers, which is what makes the next idea possible.
That idea is network slicing. Imagine carving one physical 5G network into several logical networks that share the same towers and fiber but behave completely differently — like splitting one office building into separate, soundproofed suites. One slice can be tuned for ultra-low latency (a factory's robot arms), another for massive numbers of tiny, sleepy sensors, another for plain high-bandwidth phone video. Each slice gets its own promised behavior, enforced down in the network. This is the same quality-of-service thinking from the QoS rung, made programmable and sold as a product.
Slicing pairs naturally with edge computing: instead of sending every request all the way to a distant datacenter, the operator runs small servers right at the tower or the network's edge. A self-driving car or a multiplayer game then talks to a computer a few kilometers away, not a few hundred — and since we just established that you cannot beat the speed of light, moving the computer closer is one of the only honest ways left to cut latency. The next guide in this rung goes deep on the edge and the things that live there.
What 6G is — and how to read the hype
Roughly every decade brings a new cellular generation, and 6G is the one now being researched rather than deployed — honestly, it is mostly a research program and a wish list, not a thing you can buy. The recurring themes are real, though: even higher frequencies (reaching toward terahertz) for huge bandwidth over very short range; integrating LEO satellites and ground towers into one seamless fabric so a single device roams between them; and machine learning for networking baked in, where the network predicts demand and tunes itself instead of being hand-configured.
When TCP meets the sky (and the bill)
Putting the access network in the air resurfaces an old, honest problem from the wireless rung. Classic TCP treats every dropped packet as a sign of congestion and brakes hard — a careful driver who slows the moment a wheel slips. But on a wireless or satellite link, packets are often dropped by radio interference or a fading signal, not by a full queue. TCP misreads that loss as congestion and needlessly throttles itself, which is the heart of poor wireless TCP performance. The fix is to let the transport distinguish loss-from-noise from loss-from-congestion, which is exactly the kind of smarter signal that modern transports and the QUIC you met in guide 1 are designed to use.
- A LEO satellite races overhead and your dish locks onto it; your packets ride up, across the constellation, and back down near the server.
- Minutes later that satellite drops below the horizon, so the network hands you off to the next one rising into view — ideally so smoothly your video call never notices.
- If a packet is lost to a passing cloud or interference, a smart transport must not mistake that for a traffic jam and slam on the brakes.
- Meanwhile the operator watches its energy bill: every satellite and every tower burns power, so where to place computation and how to idle the radios are now first-class design questions.
That last step points at a frontier that is easy to forget: network sustainability. Thousands of satellites must eventually de-orbit and burn up rather than become space junk; millions of cell radios and edge servers add up to a serious share of global electricity. The genuinely modern question is no longer only "how fast can it go?" but "how many bits per joule, and at what cost to the sky?" A network engineer in the coming decade will be judged on energy and orbital footprint as much as on throughput.