Announce broadcasting is unreliable

[jagerman]

As expected, throwing a random component into whether nodes forward announces is causing massive inconsistencies across the network. Different nodes have considerably different views of what nodes are online. For example, this is just two:

Uptime counts:
Node 1: 123 nodes <= 2m; 233 <= 10m; 95 <= 30m; 41 <= 60m; 48 offline (>1h).
Node 2: 81 nodes <= 2m; 146 <= 10m; 142 <= 30m; 79 <= 60m; 83 offline (>1h).

The entire rationale for 1-minute (originally; now 90-second) announces was that it is important to know when a node goes offline as soon as possible.

That is not even remotely close to being accomplished here; even a 1-hour threshold is resulting in a large number of false positives: node 2 above hasn't heard from 17% of the network in more than an hour.

A short announce frequency with flakey-by-design announce propagation through the network is, to put it mildly, a bad design.

[rainmanp7]

He's talking about the various servers not communicating all the same.
In referring to the announcement time given off and response from other nodes.
But not realizing people setup there nodes differently.
Some people are behind different hardware and specs.
I'm still wondering if he is pissed at the random reward given out..
That was my own personal concept a long time ago ;-)