Networks
Let every store of knowledge reach every person.
Everything for this step
- Field guide
- Computer science → 6 sections
- Builds
- None yet.
- Reference sheets
- None.
- Sciences it rests on
- Computer science
- Track
- Electronics and computing, level 1 of 1
Why this matters
A library helps whoever is standing in it. A network makes every library available everywhere at once, which is the largest change in the distribution of knowledge since printing — and it makes the difference between scattered settlements and one civilization.
The shape of it
- Start with what you already have. Radio schedules, a postal service and a telegraph network already move information across distance. A network is an upgrade in bandwidth and latency, not a new capability.
- Design for partition, not for perfection. Links will fail. Store-and-forward — where a message hops between nodes and waits when a link is down — is far more robust than requiring an end-to-end path, and it is how early networks actually worked.
- Publish the protocol openly and keep it simple. An open, documented, boringly simple protocol will outlive every clever proprietary one. This is a lesson the first internet learned expensively.
- Route around failure automatically. No single node should be load-bearing for the network as a whole.
- Replicate the archive at every node. Bandwidth is scarce and disks are cheap; a full copy of the library at every settlement is more robust than a fast link to a central one. Design for the network being down, not up.
- Decide governance before it matters. Who may connect, who resolves names, who can cut someone off. These questions look administrative and are actually constitutional, and they get settled by whoever moves first if you leave them open.
- Watch what it does to attention and to truth. The network you came from distributed knowledge magnificently and also distributed falsehood faster, concentrated attention into a few hands, and made shared reality harder rather than easier. That was not inevitable — it followed from specific design and funding choices. See Patterns.
Doing it, step by step
The summary above says what has to happen. This says how, with the numbers that decide whether it works.
1 · Start from the links you already have
Radio schedules, a postal service and a telegraph already move information across distance — see radio and telegraph. A network is an upgrade in bandwidth and latency, not a new capability.
Measure what you have: messages per day, delay, and reliability on each link. Then add capacity where the measurements show the bottleneck.
2 · Design for partition: store and forward
Links will fail. Design for the network being down, not up.
Store-and-forward: each node holds a message until it can pass it to the next node, which does the same, so a message crosses the network without ever needing an end-to-end path at once. Every message carries a unique identifier, a source, a destination and a time, so duplicates are discarded and delivery can be confirmed. This is how early networks actually worked.
3 · Publish a simple, open protocol on paper
An open, documented, boringly simple protocol will outlive every clever proprietary one. Write down message formats, addressing, error detection and retry rules, print it — see printing — and distribute it to every node.
Give it a version number, and require new versions to accept messages from old ones. Keep a reference implementation's source on paper in the archive.
4 · Make no node load-bearing
No single node should be able to take the network down. Give every node at least two links to different neighbours, and route around failure automatically — each node keeps a table of which neighbours reach which destinations, updated as links come and go.
Test it deliberately: disconnect the largest node and confirm messages still flow.
5 · Replicate the archive at every node, and verify it
Bandwidth is scarce and storage is cheap. A full copy of the library at every settlement is more robust than a fast link to a central one.
Keep a checksum for every file in a signed index. Nodes compare checksums and exchange only what differs. Re-verify every copy on a schedule, since storage corrupts silently, and repair from another node when a checksum fails.
6 · Decide governance before it matters
Who may connect, who resolves names, who can cut someone off. These look administrative and are constitutional, and they get settled by whoever moves first if left open.
Write into published rule: the admission rule, how names and addresses are assigned and by whom, the grounds and procedure for disconnection with an appeal, and how the protocol itself is changed. No single operator holds all of these powers.
7 · Build provenance and correction in from the start
The network you came from distributed knowledge magnificently and distributed falsehood faster. That followed from specific design and funding choices. See Patterns.
- Every document carries its source, date and author, and where possible a signature that proves it was not altered.
- Corrections are linked to the original, so a reader of the original sees them.
- The archive keeps every version, so what was said and when can be checked.