Re: european servers.. (fwd)

[Chris Behrens <cbehrens@xxxxxxxxxxxxxx>]

On Thu, 30 Oct 1997, Christophe Kalt wrote:

> On Oct 29, Mikael Abrahamsson wrote:
> | Before (this might have changed the past year or so) the L-lines seemed to
> | take effect at a very late state, when the whole connect burst was
> | completed. Is this still so?
> 
> The way it is, the servers send the burst no matter
> what. (well, as long as you have valid password and CNs ;-)
> 
> servers are sent first as you probably know,
> 
> so on the L line server side, the burst is stopped before
> any damage is done (as soon as a server matching the L line
> is found, the link is dropped).

I disagree =)  I think certain damage is already done by sending
some "SERVER x"'s back into the network before the server is disconnected
for a SERVER that matches an L: line.

What we'll be doing on EFNet is TSing the SERVER messages, I believe,
although, maybe we should just "hold on to" SERVER messages until
we get a non-SERVER message =) ...and then at that time, send them in
1 big lump to the other servers connected to you.  Hmm...I suppose it
is possible for there to not be a message after 'SERVER' though, if
there are no clients connected to the server you link in.
If this idea actually worked, though, you could probably call it
Server-Delay...thus, it would fit in perfectly with IRCNet ;)

- Chris


> 
> on the other side, the burst is processed as quick as
> possible (that's what we all want, right? ;-) and since
> there's no error, users are broadcasted, collisions happens
> or not.. until the link is broken by the other side.
> 
> I don't think the behaviour has changed in a long time.
> 
> | > this sounds like some kind of "reverse" L line.
> | 
> | Yes, but it is nice since ot puts me in control of which servers I want to
> | connect to when other servers are online. L-lines have to be changed by
> | the party at the other end, meaning that if I have 4 different uplinks I
> | have to get all of them to implement the same L-lines if I get a 5th
> | uplink.
> 
> yeah, which shouldn't be a problem in theory, but is in
> practice given the lack of and lag in cooperation which
> characterizes all big nets.
> 
> | > also, the probability of you connecting to the new born
> | > server before it connects to its normal uplinks is rather
> | > close to null, isn't it?
> | 
> | Same thing happens when one of my uplinks split from the rest of the net
> | and my server tries to catch it. I think this situation happens often
> | enough to warrant some attention.
> 
> good point, new born servers are probably not an issue, but
> freshly split server are likely to be.
>