Decentralized Computer Networks

1998-10-14: I start this page

1999-08-03: I add ACN

1999-10-10: I add Rooftop

2000-05-11: I add the "Parasite Network" article

Several days ago, there was an article on Slashdot about an interesting story on EE Times. Basically, the idea is this: give lots of people short-range, low-power radio transceivers they can connect to their computers. Then, they can talk to each other over long distances, by routing their messages through this network of radio transceivers.

The beautiful aspect of this network is that it arises out of people cooperating with each other for mutual benefit, just like society, the World-Wide Web, the market, marriage, the Linux development project, and any number of other important forms of human organization. This is very different from the command-and-control corporate model, wherein owners (shareholders, founders, etc.) employ employees to provide services to customers. Excellent though that model is for many things, I think the basic bedrock of society should never be run that way.

The Slashdot comments had pointers to many similar projects, and there are some other projects I know of, too.

• The L0pht's radionet project
• The MIT Media Lab has toyed with something called Hyphos, which is intended for hundreds or thousands of devices in a small space (like a living room or a house), as part of their Things That Think project.
• Metricom's Ricochet is an operational system that works along similar principles; they have boxes mounted on light poles all over the metro area that forward your packets on toward their destinations. The over-the-air data rate is 100kbps, but typical transfer rates are more like 28.8kbps. They have ``Wired Access Points'' that provide access to the Internet. Their modems have a range of about 1,000 feet, and their repeaters have a range of half a mile to two miles. They operate in the 902-928MHz band, at emissions levels below those that require a license (under Part 15 of the FCC rules.)
• DIRC (formerly at http://www.dirc.cetecom.de), a German organization founded by Winrich Hoseit. It proposes to do things much as has been described above, but charge about US$25 a month to each subscriber. It says today that it was founded two years ago, so late 1996. There are several online documents about DIRC: a Computer Club Report, an article in Gateway magazine, and a few others, all in German. There's a Malaysian home page too. Each channel gets something like 40kbps, and there are 640 logical channels, with a range of 5 km. The technology is currently seeking two PCT patents: 2 Aug 1997 Nr. 197 33 586.1 and 30 Aug 1997 Nr 197 37 897.9. The transmission is between 700MHz and 2.4 GHz. Each station transmits up to two watts. Each station can supposedly get 40Mbps.
• There's a paper on munchkins that discusses ``decentralizing routing, naming, and accounting services to an unprecedented degree to enable ubiquitous, untethered computing''. A munchkin is an extremely cheap, extremely low-power communications device about the size of a golf ball that will be manufactured in the trillions to blanket the Earth.
• The RUNES project is working on network protocols that could conceivably work on such a network.
• Packet radio has been doing this for years, at ridiculously low speeds, and with a variety of political restrictions. The ARRL has an introduction to packet radio online. They use a protocol called AX.25, based on X.25. Quote: ``I don't think any knowledgeable person would argue that AX.25 is even close to being an efficient protocol for HF work.''
• SFLan is trying to do the same thing --- cooperating users building a network. They've gone live in the Presidio in San Francisco.
• Airborne Communications Node is a proposal to build a Ku-band-and-satellite-and-some-other-stuff self-organizing network backbone for voice and data. It would be used by the US military.
• isen.com's SMART Letter #28 describes a startup called Rooftop Communications, recently bought by Nokia, that works the same way. The article cites "DSL speeds", a range of "up to a couple of miles", line-of-sight transmission, several beta nets in the SF Bay Area (not clear if this means Rooftop nets or Ricochet and SFLan), pilot nets in Morocco and "Guinea", node prices from $2000 to $5000, 50-ms-per-hop latency, a proprietary OS called "IROS", meer.net as a Rooftop-capable ISP in the Bay Area (although it doesn't mention this on its web pages), and a few other details. No links to other information are provided.
• Let's Build the Parasite Network, says Jim Griffin, suggesting the same thing all over again.

There were several silly objections in the Slashdot comments; for completeness, they are

• No money will be invested in maintaining the communications infrastructure, so it won't improve.
• The FCC won't approve it.
• The hardware isn't free.
• It's not secure.
• It doesn't guarantee bandwidth.
• What if my neighbors won't forward my packets?
• Nobody will buy one.

These arguments are so obviously ridiculous it's not worth it to respond to them; if you can't see why one of them is ridiculous, email me and ask for an explanation, and I'll write you one.

There are some legitimate questions, most of which came up in the Slashdot posts, though:

• Forwarding many times means relatively high latency. Packet-forwarding latencies will have to be nanoseconds to microseconds, not milliseconds, to get decent latencies.
• Routing is difficult.
• There may be difficult political problems in many countries.
• Routing is even more difficult in the presence of possibly malicious nodes that may lie about their identities and connections.
• Reliable transmission is difficult when some of your gateways may maliciously alter your data.