[PATCH] Maintain outgoing TCP meta buffer

[Scott Lamb]

On Jan 17, 2006, at 5:14 AM, Guus Sliepen wrote:
> I don't want to change the meaning of already existing configuration
> variables.

Well, I'd like to split the meaning of PingTimeout into two:

* how often to send pings
* how long to wait for responses

because I don't think it makes sense to configure these to be the  
same. Especially since the next item in my tinc to-do list is to have  
a way to force pings to happen immediately upon network changes. I'd  
like to open my laptop and have tinc realize in five seconds that its  
existing TCP tunnels were opened on a totally different network and  
are thus invalid. I think it's reasonable to expect responses in that  
time. Yet I don't want to cause extra packets to be sent every five  
seconds.

My first patch kept PingTimeout meaning the first one; my second  
patch kept it meaning the other. Would you prefer to introduce two  
new options? I suppose it could issue a warning if the old one and  
either/both of the new ones are specified. So "PingTimeout" would  
give both. "PingInterval" would be the first only; "PongTimeout" or  
maybe "PingResponseTimeout" would be the second. I like the way of my  
second patch, but I'll go with whatever you prefer.

> Also, I still think it is better to put a limit to the size
> of the buffer, and drop PACKETs when the buffer is larger than that
> limit. Late packets rarely do any good anyway, best to get rid of  
> them.

Sure. That's not hard, and you're probably right. But I'd like it to  
be in addition to the ping changes, not instead of them.

--
Scott Lamb <http://www.slamb.org/>