[Cluster-tech] PC41 locations and things
Dirk Koopman
djk at tobit.co.uk
Sun Jan 14 00:30:33 CET 2007
Lee Sawkins wrote:
> Dirk
>
> I would like to try and use these PC9x messages.
> What does spider require from my CC Cluster software to enable them?
>
Lee
At the moment I am struggling with the real problem area I am trying to
address, viz: routing. The PC41 replacement stuff will come after I have
made that work completely reliably and then added the universal
talk/chat/announce replacement sentence.
It mostly works. However, there is an annoying edge case that I am
trying to track down which causes routing information to disappear on
for locally connected nodes. I also have some more work to do with
automatic route discovery and related issues. However, here is a summary
of how it works:-
1. The biggest change is that nodes only (ever) send their configuration
(using a PC92) and the configuration of any *directly connected*
"traditionally routing" nodes (TNODES) (ie still using PC16,17,19,21).
The config stored of any TNODES is limited only to the local users on
those nodes. Anything else is kept as hints, but is not transmitted
onward, neither to other PC9x nodes nor other TNODES. The only
configuration that other TNODES will see are the composite of all the
PC92 nodes's configs + any other locally connected TNODES.
2. PC92 Nodes periodically output their configuration (currently every
every 30 mins). Failure to receive a config after 3 update periods will
cause that node's config to erased (and the changes to be propagated to
any connected TNODEs). Think OSPF.
3. Although I am doing some Adds and Deletes, this will cease and the
only routing updates will be periodic configs. Hopefully this will
reduce some of the bandwidth.
4. All PC9x sentences contain a timestamp and the originating node call.
This allows any of these sentences to be deduplicated and deliberate
loops (for routing and new other functions) will allow things to
continue to work as only "new" PC9x sentences will be processed.
5. All PC9x sentences are passed onto neighbouring PC9x unaltered apart
from a (now more reliable) decremented hop count.
The point of these changes is to make nodes responsible for maintaining
their own routing table rather than trying to interpret the true picture
from the constant noise that comes in from outside which really is only
someone else's view - which is, we have seen, wrong.
The consequence of this is that certain things, like mail, will need a
polling mechanism (rather a "who has" route enquiry) which is slugged to
happen no more often (at least automatically) than the route update
period (as mentioned earlier: 30 mins). Talk/chat/announce has a
slightly different approach in that, if it is a talk, and there is no
route, then it will be broadcast, much like chat/announce.
I will let you all know when this comes out. But it is actually being
actively tested on three network nodes and I hope to have a beta out in
the next week or so.
Dirk G1TLH
More information about the Cluster-tech
mailing list