Misunderstanding of Subnet directive in hosts files

[SVM]

> Hm, that is strange, once a connection has been made through NAT it should keep
> working, unless the connection timeouts on the NAT device are set very small.
> You could try setting "PingInterval = 30" in tinc.conf to have tinc send ping
> packets more often.

It doesn't help, unfortunately. Works only if TCPOnly=yes.
Or if I say ping <rootnode-ip-addr> via tinc tap interface to "root" 
node, but also just for approx. 1 minute. It's definetly due to double 
NAT, I tnink.

Here is one more question. How does "internal" Ping between nodes go? 
Via meta- or data-connection, just like usual icmp-echo-request?


>> If I leave TCPOnly=no(default) tinc cannot determine itself to use
>> tcp instead of udp in my case as it described in documentation.
>
> Tinc should determine it itself. If it does not, that is a bug. It can take a
> minute though for tinc to detect that UDP has failed to work. Does it still not
> work for you after more than a minute?

Yes, still doesn't work. Maybe here is situation, when at first tinc 
works fine via UDP, but after a minute - there is no more translation 
rule on NAT-device of my provider? Should tinc do this check only at 
start or somehow else?
I use tinc 1.0.11-1 from ubuntu 10.04.4 universe repo.


> They will be in one TCP session.

My "root" node has two uplinks from different ISP and I want client 
nodes could make two connection to the same tinc-daemon on "root" but to 
different ip-addresses for redundancy purposes(in case of one uplink 
will fail).

May I write two "Address=" lines in one host file? or simply use two 
different host files for each connection?

I was trying to use two host files, but I can't see the second 
tcp-connection and I don't know if tinc will switch to the second one if 
the first will fail in any case.


Very appreciate your help, Guus.