[Uk-dxcluster] Network loops again

[uk-dxcluster at dxcluster.org]

On Sun, 2005-02-27 at 21:14 +0000, uk-dxcluster at dxcluster.org wrote:

> Re: the network showing a true picture of nodes' connections, 'DXC is 
> currently connected solely to GB7BAA.  I've just had a look on several other 
> nodes in the network; 'BAA and 'DJK and a few other nodes show our 
> connection correctly but 'YDX is showing 'DXC as being at 'MBC, 'DJK and 
> 'UJS!!!  Methinks there's something not quite right somewhere along the line!

The PC protocol is not capable of supporting the transmission of
configuration information correctly in a looped environment. In fact I
would argue that it is inadequate, or at least ambiguous, in a
non-looped environment of more than one hop.

So after several false starts, much hubris and delay, I have been
working on a new protocol which is called "Aranea". It is successfully
transmitting reliable routing information on a deliberately looped,
mixed PC and Aranea, network to both Aranea controlled and directly
linked PC protocol nodes. It is currently running on four nodes (GB7TLH,
GB7DJK-1, WR3D and GB7MBC-1). It is transmitting routing information and
also moving DX Spots.

As time permits over the next month, it will have more functions (ann,
talk, wwv etc) moved over from PC protocol to Aranea centred control. In
time, the Aranea view of the network will become the dominant,
controlling, view (amongst DXSpider nodes) and the stuff coming in on PC
protocol (from nodes not directly connected to an Aranea node) merely as
hints as to the possible location of a user or other node. This will
mean that the nearer (in hops) you are to an Aranea node, the more
likely it is that the configuration that you see will be accurate.

Note that the major reason for this, is that Aranea works by making sure
that, with the exception of directly connected PC Protocol nodes (for
whom it acts as a first level proxy), each node only transmits routing
information for itself. Nodes do not pass on the lies passed up to them.

Also broadcasts are consumed in "pump up integrators", on receiving
nodes, so that after a period during which 3 broadcasts could have been
received, if none are, then that node is deleted. One broadcast received
is enough to make it live for a possible further 3 periods.  Rather like
modern(ish) routing protocols such as OSPF et al.

Nodes that have attached PC protocol nodes will still see the network as
it is presented down those PC protocol interfaces, except that
configurations of nodes that come down the Aranea ports, will superceed
(and modify by issuing appropriate PC protocol) that which has arrived
thus far by PC Protocol. 

Because the Aranea broadcasts repeat over a period, and because those
broadcasts compare the routing table with the broadcast and then send
adjustments in PC protocol, this (already) has the effect of steadying
the view seen by attached PC Nodes. Thus causing them to be more
convergent with reality. As more Aranea nodes come on stream and connect
to each other, so this effect will spread.

What I hope will happen, over time, is that the loops will be
transferred to the Aranea protocol network and PC protocol connections
will go back to being one or two level, non looped trees. Hopefully then
it will all stabilise.

The result will be DXSpider Version 2.

Dirk G1TLH