FindGroupResponse without Sentinel has bad scaling behaviour

[benjaminbollen]

Without Sentinel FindGroupResponses come back:

for n < 32 connected nodes in the network: this is n FindGroupResponses, each with n nodes listed;

So node n+1 on connecting to its group sends out n^2 connect requests; each of which are swarmed over the n-nodes. This is n nodes sending it to n-1 other nodes.

so every new node n+1 has n^2 * n * (n-1) = n^4 - n^3, but have to be handled on the same machine
2 -> 8
3 -> 54
4 -> 192
5 -> 500
6 -> 1080
...
15 -> 47250
...
32 -> 1015808 message handlings for a single node to connect

[benjaminbollen]

merged; improved performance