[Dxspider-support] Setting maximum connections

Dirk Koopman G1TLH gb7tlh at dxcluster.org
Fri Oct 3 12:56:47 BST 2008

Dirk Koopman G1TLH wrote:
> To summarize:
> * If you connect to them (ie you have a connect script and you either
> manually or automatically connect via crontab), then you need do nothing.
> * If a node connects to you (via your listen port which is usually 7300 
> or 8000) then you need do nothing UNLESS that node already has 8 or more 
> other connections to other nodes. I call these "big" nodes.
> * If a "big" node connects to you, then either you need to raise the 
> default $main::maxconnect_node to a value larger than 8, or disable this 
> feature completely: by setting it to 0 (using a set/var in the startup 
> script), or individually set up an exception by doing a
>   set/maxconnect nn <node call>.
> And to remind people, this only applies if you are using 1.55 build 
> 0.30. And to remind other people that if you are using 0.28 or 0.29 you 
> should upgrade to 0.30.
> If you are using 1.55 build 0.27 or below you can ignore all of this.

I should add that if you are a "big" node and you connect outwards to 
another node, then you will need to get them to do set/maxconnect nn 
<your node call> for you on THEIR node.

The point of this is twofold:

1. To discourage users from having more than 2 or 3 connections which is 
simply a waste of bandwidth.

2. To encourage nodes to interconnect more widely than is currently the 
case. If we want to have a resilient and fast network, then we need to 
stop relying on a few "big" nodes. If one of these crashes then it can 
have a disproportionate effect on the rest of the network.

Also running a big node implies a greater bandwidth requirement, which 
can be expensive (especially here in the UK).

The ideal, in my view, is for every node to have between 3 and 5 partner 
nodes (preferably with a big geographical spread) and say 10-50 regular 

But I am in no hurry to achieve this and it may never happen...


More information about the Dxspider-support mailing list