Decentralizing the Internet So Big Brother Can’t Find You

[J.A. Terranson]

Note: Reply made on a free form stream-of-consciousness basis.


On Fri, 18 Feb 2011, Eugen Leitl wrote:

> If you're talking mesh, one of the basic requirements is to get
> rid of central address allocation authority and establish
> a local-knowledge based routing. 

Thats not as easy as it sounds, especially in hyperdense cities - although 
I will also admit that this is more a 70% administrative issue than an 
technical issue.

It's unlikely that anyone in their right minds is going to just open up 
their networks to both unlimited and unmonitored connections, but even if 
they do, they are going to demand some allocation control and 
accountability.  Not that it's necessary for any technical reason, but 
because it's required for the people who both pay the network bills and 
receive FBI "letters" / subpoenas: even the smallest networks have some 
kind of "security department", even if it's the same guy who does 
everything else there.  This is where the idea truly breaks down.  I 
wouldn't allow an open wifi connection when I know that any illicit 
activities are going to come back to haunt *me* when the FBI does their 
next "predator sweep".  No municipality is going to allow geocoded/derived 
addressing for the exact same reason: they want to be able to point the 
finger when the time comes.  They also want to know if your node is acting 
up, where they can go /phone to make you fix it.  

Mesh networks are a political disaster *before* anyactual engineering time 
goes in.  But if you are lucky enough to get past the politics (hard.  
really hard.), then you start down the actual engineering pipe: traffic 
planning and engineering is at best difficult, at worst impossible, 
without centralized control. So do we just assume that the warts know not 
to exceed bandwidth(x), and plan for max(y) warts per square 
mile/kilometer?  Do we assume that all these warts are going to be using 
passive connections for ftp?  this road leads back to political questions 
by the way. 

The residential carriers will go postal if 100k home connections are 
suddenly offering services and draining back through the cable/dsl drains, 
against their AUP/ToS. If you choose to create layer 2 drain solutions you 
are back to tight and centralized allocation control.

> If you get that far, the human operator mistakes are removed from the 
> loop.
>  
> > I was with a group that tried to work out a mesh implementation across a 
> > relatively small (~15sq miles) area, and it never came to fruition, 
> > despite several years of work on the problems presented.
> 
> Do you have a pointer to description of your project, and what went
> wrong?

This was a municipal project, so I can't point you anywhere: sorry.  As 
for what went wrong: everything.  Everyone who touches the mesh has 
questions/issues of the nature I alluded to above.  Muni IT demands to be 
able to track back every ip to a known user, meaning any kind of 
geolocation solution is gone right off the bat.  By the way, the reasoning 
here wasn't what I was expecting: they were more concerned with virus 
control and quarantine (which implies a separate layer 2 sandbox for 
infected connections, or connections carrying infected traffic where such 
can be accurately and reliably determined - which is about 50% of the 
time).  The muni wanted a known IP for each user, with certain wifi 
hotspots exempted to local rfc1913 networks using DHCP.  We had 
interesting times with these hotspots too!  Not having enough IPs on a 
public wifi segment causes weird behaviour that the basic call desk cant 
troubleshoot, and that they don't recognize as being broken - so these 
spots get reputations as being "unreliable".  Significant automation was 
assembled just to keep track of required resources vs actual available 
resources, with traps going out when automation sees something we dont. 

The traffic engineering issues are perverse.  Meditate on them for a 
little while, and consider all of the ways you can slice your own throat.  
Is a steady rise in users a trend, or did the bookings for conventions 
just have a good year?  Are you meeting peak demand, or are you 
oversubscribing on your backbone?  *Is* there a backbone?  Lots of 
questions at layers 1 and 2!  In fact, layer 2 is where you're going to 
spend a *lot* of your time, unlike the operation of a more traditional IP 
network, where you spend the vast majority of your time at layer 3.  
Segregation is a necessary ingredient, as well as scanning-on-allocation 
to check for blatantly rooted connections.  This gets tricky as hell.

We were asked to plan for each muni supporting drains for the connections 
made within their geographic areas: oops! Looks like I want geo-derived 
addresses again!  Except that these same munis have the additional 
condition of only wanting to allocate for their own taxpayers: oops!  Back 
to centralized control and allocation...  Round and round you go...

Everyone here knows IP well, and can add a hundred things that I've not 
touched on.  Every time you add a condition, you increase complexity, and 
these increases are *not* linear!

> I can tell you one thing: Eben Moglen is not a stupid man, and he's

I never alleged that he was stupid.  I know he's no moron.  But he's also 
not thinking this through, he's looking to build the device before he 
knows his application.  Thats fine for general purpose machinery with lots 
of cycles and storage to spare, but a wart is likely to be max'd out at 
ultra-small storage and cpu limits:  virtually every wart-driven device 
was designed to act as a simple bridge or router, which requires 
surprisingly little computing power: An old 386 is massive overkill for 
acting as a router on lines up to about 3mbps - while these warts may have 
more *cycles* available than an old throwaway box, they are going to have 
[at best] a few hundred megabytes of flash storage: not a lot of room 
there for anything meaningful.

> surrounded with technically capable people. If he can raise the money
> for 10-100 k wall wart units you can assume people will do useful
> things with them.

I can assume people will do *something* with them, but *useful* things, 
no, I cannot just assume that - that will be something I'll have to see to 
believe.

//Alif

-- 
"Never belong to any party, always oppose privileged classes and public
plunderers, never lack sympathy with the poor, always remain devoted to
the public welfare, never be satisfied with merely printing news, always
be drastically independent, never be afraid to attack wrong, whether by
predatory plutocracy or predatory poverty."

Joseph Pulitzer, 1907 Speech