Full-Mesh and lots of nodes - Re: random problems with Forwarding=off

[Sven-Haegar Koch]

On Fri, 16 Mar 2012, Guus Sliepen wrote:

> 64 nodes is not an unusual number for tinc. In what way does the network become
> unstable?

In our network we had more or less the same problem.

Reason were (as far as I could see) offline nodes, and (dyn-) DNS 
lookups, which for some nodes may take a while - and while the lookup 
is in progress tinc does not do anything else.

With a few nodes you do not feel it, but with lots of them there were 
noticeable delays resulting in packet loss. (Just a small delay per 
remote node, but with >100 nodes you have this delay nearly always)

We changed our topology to have a few (5-10) "primary" nodes where 
everyone connects to instead of full-mesh, and since then do not feel 
this problem anymore.

Data transfers still go peer2peer where possible.

c'ya
sven-haegar

-- 
Three may keep a secret, if two of them are dead.
- Ben F.