Re: Munchkins

From: Eugene Leitl (
Date: Fri Mar 09 2001 - 02:11:23 PST

On Fri, 9 Mar 2001, Danny O'Brien wrote:

> On Fri, Mar 09, 2001 at 07:45:10AM +0100, wrote:
> But people can hack together the most amazing solutions based on the shittiest
> beginnings. *That's* biped evolution.

Don't get me started on frozen chance in evolution. Limited choice of
materials, ad hoc bauplan recycled in infinity, molecular shenanigans
we got stuck with just because interstellar material is lousy with them.

We humans tend to do the same, though with the benefit of being able to
cross relatively sterile fitness spaces by virtue of modelling it in
the virtual sandbox between our ears. But we're prone to notoriously
overestimate our facilities in this are.

The more reason not to fuck up the foundations, now.
> The reason I pushed out the MobileMesh protocol onto FoRK last week, is
> because you're going to start seeing people rolling out hacked-together 802.11
> community networks in the next month or so. All of the solutions are

So it has started? Great.

> improvised, none will be perfect. Something will prevail. Physically, 802.11
> is the barest minimum you need to get munchkinesque system working.

The 802.11 only does allocation by frequency slices and no realtime
beamforming (active smart antennas), right? Can the hardware use several
frequency slices simultaneously?

> Prices of units are collapsing: to the OS, it looks like Ethernet. It does
> Ad Hoc peer to peer out of the box. People can see the advantages -

I only need DHCP enabled when walking around, right? Does the OS layer
deal with changing IPs transparently?

> particularly in the recent post-Napster climate, and the slow crawlout
> of 4G wireless and broadband, of building a local decentralised Net.

Thanks for the report from the trenches. I missed that it has started

> I'm trying to work out given that what the minimal system is that can grow
> into the system you're describing. If you want perfection, you're fucked:

The design needs to be flexible enough to accomodate future growth without
hard breaks, inducing incompatibilities. If you need input from the
lunatic fringe (hey, I happen to live here), I'll be glad to contribute.

> but you can slip in some values now that will at least let people bolt on
> better solutions in the future. Not only is that a better solution than
> whining about how no-one is using your Godly perfect crypto-agora masterwerk
> (available RSN), but given that you don't know the future either, it may even
> turn out to be more flexible.

I agree with you that whining is pointless, so within my time constraints
I'm willing to collarborate. Given the connections, FoRK should be able to
marshal a number of good architects. I suggest we take this off-list.

> So, say you've got 500 people, in a bunch of cities around the world, who are
> interested in growing a decentralised, mesh-style network. They're going to do
> it, whether you like it or not, and far too soon. Even if it's a sucky system
> they build - like Gnutella, say - they'll be people trying to fix it for
> an eternity, and doing better than you could possibly imagine. What would
> you set them up with? What's the minimum that you can get away with? That you
> can put together in a few weeks?

I have no clue, but currently you usually have a local cell around a node
connected to the Net via xDSL. Isolated nodes can link up via a VPN. Where
cells overlap, you need routers, which have several wireless NICs, and
possibly directional antennas. Incidentally, LinuxBIOS (where you flash a
kernel into mumboards native flash or disk-on-chip instead of legacy BIOS
zombieware) is getting ready for end users. Initially, those need to be
manually configured, before one can figure out how to make routing tables
set themselves up automatically. On top of this you can run your
sharing/routing/remixing system of choice, whether Freenet, or MojoNation.
It would make sense to add a lot of memory to the routing nodes, and
install big ramdisks in them (HDs are less suitable, since you'd want to
put the routers in weatherproof boxes on people's roofs).

> I figure it's a downloadable userland dynamic routing system for Linux, with a
> clear possibility of being ported to Windows. That gets them wireless

Initially, yes. Later, you have to create reference software radio designs
based on embedded PCs and true embeddes, and exploit options like digital
pulse radio.

You can estimate distances from signal strength (if your hardware
gives you access to the innards) which is probably correlated with
channel throughput (dropped packets, and the like). Until we'll
get true mutual time-of-flight measurements, I suggest allowing
people to initialize nodes WGS84 coordinates. That knowledge should
be strictly local (i.e. direct and next neighbours).

> routing (what they want). On top of that, each client generates a
> keypair, and dumps the public key somewhere public. You associate the key and
> the client to an allocated 10.*.*.* IP address, just so you can allow some
> sort of mapping to existing TCP/IP. You put in a MOTD feature, so
> people know when there's something better to upgrade to. You hack the routing
> protocol to allow dispersal of the location of these few elements of central
> servers (the MOTD, the IP allocating server), so everyone's not fucked when
> the central servers move. You send this out, then work on the better,
> decentralised version.

I need to think what you're saying here.

> I'm not a munchkin expert, so I don't know whether that's enough, or too
> little. But I do know that at every cypherpunk meeting, here at CFP, at MIT,
> on forums and at hacker meets, I hear 802.11 mentioned about once every
> fifteen minutes. Something's going to blow.

Cool. I hope it will be a big one, and they'll reach criticality before
you see lots of cars with rotating antennas on them driving through the
> What would be your minimum roll-out?

Are we talking purely grassroots, or activists which assemble hardware
and sell kits?

Once again, I suggest we take this off-list. If there's no suitable
channel, let's go to yahoogroups, or equivalent.

This archive was generated by hypermail 2b29 : Fri Apr 27 2001 - 23:13:44 PDT