[Dxspider-support] Old routing info. What to do ?

[Dirk Koopman]

On Fri, 2003-01-17 at 19:51, Rene Olsen wrote:
> Hi.
> 
> There has been a lot of talk about routing problems lately. I have done a bit of 
> investigatiom, and I think that the result is not very uplifting.

You are telling me! 

> But, the mistake had been made and now after 1-2 months that old info is still floating 
> around the net. Really amazing that it can be possible.

True... That is because at no time is the cluster now completely
disconnected from itself long enough for the nodes to disappear
completely. It can take over an hour for a node to actually disappear
totally from everywhere in the cluster once it is gone. 

At least part of why this happens is because, although you see maybe 700
nodes (as I do regularly) there are many, many more. These 'silent' (to
you) nodes are the 'reservoir' of old routes. This is partly why I was
always so dead set against passive links.

In fact it probably isn't just clx related. It is more to do with the
latency of the network joining cluster together. 

This is why you can get announce/pc41 storms that cycle round every 20
mins or so.

> 
> Now, what do we do about this. It can't be that we have to live with this. We need to 
> figure out a way to fix this.

At the moment I have created a branch in CVS and am now going thru the
code doing some serious rewriting (on that branch) to allow me to
support two protocols at once. NP is slowly emerging from this mess (and
it *is* a mess, currently). 

It is becoming clear, whilst I am doing this, that it is likely that I
shall only allow 'full protocol' with spider / compatible nodes (if any)
and will, at least to start with, only send local configurations
inter-spider. 

The next stage will be to allow the PC19/16 from directly connected
nodes through. 

After that, I shall see what it all looks like. But it is likely that a
sysop will have to perform an operation to explicitly allow the
forwarding of routing information from more distant nodes - on a node by
node basis and that they can a) ping them and b) trust the data.

In fact current thinking is to completely ignore PC19s (and treat PC21s
with grave suspicion) from remote nodes on the grounds that you don't
know where they are coming from and the data is completely unreliable
anyway.

This means that many 'conventional' nodes will, for a time, drop off the
network network. For some it may be permanent unless they establish a
direct link to a/some spider node(s).

> Lets get a good debate going about this, and lets hear what Dirk thinks about it.
> 

What I think is: this will cause great wailing and gnashing of teeth.
Not least because this will become another autonomous network like the
ARC one. 

I really would not like that, but I cannot see the alternative.

Dirk G1TLH
-- 
Please Note: Some Quantum Physics Theories Suggest That When the
Consumer Is Not Directly Observing This Product, It May Cease to
Exist or Will Exist Only in a Vague and Undetermined State.