*
**
*** 1- The problem
**
*
                                                                                                          
          *******                               *******
         *       *                             *       *    
        *  GNODE  *                           *  GNODE  *                          
         *       *                             *       *   
          *******                               ******* 
                           +----------+    
                           | INTERNET | 
                           +----------+    

          *******                               *******
         *       *                             *       *    
        *  GNODE  *                           *  GNODE  *                                                        
         *       *                             *       *                                                      
          *******                               *******                                  
                                                                     
                                                                     
The problem is to join the Gnodes with internet. 

*
**
*** 2- A simple approach
**
*

Consider only two gnode, with only one vnode:

    GNODE A                        GNODE B

      *  *  (VA)               (VB) *  *
    *    *  *---      INET    ---*     *   *
     *  *                            *   *
     *    *

The vnode are named VA and VB.
Suppose also that the inet address of VA is know.

VB decides to contact VA.

*
**
*** 3- The vplama daemon
**
*

The vplama daemon is a daemon running on vnodes.
Initially, it listens on a locally unix socket and on the Inet 
interface.

The unix socket will be used to talk with ntkd.
The network socket will be used to talk with other vnodes.

For each vnode, a tunnel will be created. Then the vnode become
a vrnode.

When ntkd has to forward a ntk packet *which provenience is internal 
to his gnode*, like a tracer, it forward the packet also to vplamad 
using unix socket.
Then, vplamad forwards the packet to each tunnel, ie to each vrnode.
Then, when a remote vplamad receives the packet, it will forward it
to his local ntkd.

When ntkd receives a packet from vplamad, the packet will be 
forwarded only inside the gnode: ie, pkts coming from inet are not
resended to any vpalama daemon.

The communication happens:

        +-----------------+                   +-----------------+
        |     **VA**      |                   |   **VB**        |
        |                 |    +----------+   |                 |
Gnode --|--ntkd    vplamd-|--  | INTERNET | --|-vplamd   ntkd --|-- Gnode
        |    |    /       |    +----------+   |       \    |    |
        |    |   /        |                   |        \   |    |
        |  Unix /         |                   |         \ Unix  |
        | Scoket          |                   |          Socket |
        +-----------------+                   +-----------------+

*
**
*** 4- Useless tunnels
**
*

Vplamad will test the tunnel, killing the dead tunnel.

*
**
*** 5- Hooking
**
*

There is no hooking, just routing. Or, in the worst case, an internal 
gnode rehooking.

Let's explain.
When VA and VB join, they control the compatibility of ntk netwrok 
addresses.

Switch:

 i) network addresses are compatible
    The VA vplamad informs the VA ntkd on the new route to join 
    the addresses of the VB-Gnode. 
    The route is the tunnel that vplamd created with VB ;)
    For VB it's the same.
 
 ii) network are not compatible
    To join, one of the two gnode has to change his network.
    Suppose that initially VB contacted VA: in such case, VA
    imposes to VB to change the addresses.
    Ie, the gnode that has to change the address is the 
    connecting gnode.


*
**
*** 6- Too Many rehook?
**
*

The problem of rehooking arises.
But it's important to consider the following

 i) Internet is very diffused, so, each gnode will have a lot
    of vnodes.

 ii) When two gnodes will be vplamad connected with multiple
     vnodes, it's safe to consider stable that connection.
 
 iii) If the connection is stable, each gnode knows continuously
      the ntk-network address of the remote gnode.
 
 iv) So, it's very easy for each gnode to choose, in the case of
    internal rehook, a compatible network.
 
 v) When a group of gnodes will be connected in a stable way, 
    the vnodes of this group will never contact new gnodes.
    They already have a good net. Ie, they will play a passive
    role: new gnodes that want to join this group will do the
    active part. 
    So (see 5(ii)) stable network will remain stable, while the
    problem of rehooking is a problem for the atomic new gnodes.
 

*
**
*** 6- Entry servers
**
*

vplamad stores the list of all his vrnodes that lives at 
least 3 days.
The daemon could be contacted and send the list.

A list of reliable, fast entryservers will be stored somewhere
on the net.