scalability and bandwidth usage?

[Guus Sliepen]

On Sat, Nov 13, 2004 at 10:21:51PM -0600, Lonnie Cumberland wrote:

> Thinking about the documentation on Tinc suggests that if it is always 
> trying to maintain a full mesh topology and if there would be many 
> nodes, e.g. more than about 100 or so active at a give time, then Tinc 
> might consume a lot of resources on that single node processor along 
> with trying to maintain all 100 connections which could eat up much 
> needed bandwidth that may not be utilized.
> 
> I suspect that there should be a way to have Tinc be aware of the 
> possible connections to other nodes, but not actually make the 
> connection to a particular node via "ConnectTO" until it is actually need.

Tinc has two types of "connections". The first type is the TCP
connection made for each ConnectTo statement in a tinc.conf file. This
connection is used for the meta-protocol, it is only used to exchange
information about the presence of other tincds and the subnets they own.
Normally (unless you use TCPOnly), no VPN packets are sent using this
type of connection. Tinc will also never create more of these
connections.

The second type is the UDP tunnel. This is normally used to exchange VPN
packets. UDP is a stateless protocol, so you don't need to set up or
maintain a UDP connection in order to send or receive data through it.
So, after learning the addresses of the other tinc daemons, a tincd can
send UDP packets to all other tinc daemons without any overhead of
maintaining an explicit tunnel.

-- 
Met vriendelijke groet / with kind regards,
    Guus Sliepen <guus at sliepen.eu.org>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
Url : http://brouwer.uvt.nl/pipermail/tinc/attachments/20041114/15becf8e/attachment.pgp