IMs and group IMs (was Re: [sldev] Frequent bugs with difficult repros)

[Argent Stonecutter]

On 2008-03-02, at 10:22, Tateru Nino wrote:
> Actually, in its heydey, there were cascade failures on IRC  
> networks when the loads just plain exceeded what they could handle  
> (or what, by administrative policy, the hosts were willing to  
> handle). Loss of a node then the excess back onto remaining nodes,  
> and the whole network go dark in a relatively short period of time.

This was due to IRC's static network model. The IRC network is a  
simply connected tree: removing a multiply-connected node ALWAYS  
results in a netsplit. Usenet uses a flood-fill algorithm, and faced  
with the same load, automatically routed around problems. In fact in  
Usenet the problem is preventing traffic escaping a limited  
distribution, not keeping traffic flowing.

And in general it wasn't traffic that took an IRC node down, it was  
that the node had to go down for some other reason and there was no  
way to route around it.

But I'm not suggesting that IRC be duplicated, just that it's a  
demonstration of what's possible on hardware far slower than what SL  
is using.