[Cluster-tech] PC92 A

[Dirk Koopman G1TLH]

Lee Sawkins wrote:
> At times my cluster has over 700 local users.  Sending PC92A packets
> with IPs is going to cause a lot more backbone traffic.  From my point
> of view, the only IPs I want to see are those for users who are actually
> sending out DX spots.  The majority of users rarely, if ever send out
> spots.
> 
> Some of my users are using dialup connections.  Every time they dial in,
> they get a different IP address.  Sometimes their dialup connection
> drops, they dial back in, get a different IP and reconnect.  The cluster
> did not know they had left, but allows the reconnect and does not send
> out a PC92 for this user.  What you are proposing now would require
> these types of reconnects to update their IP info by an immediate PC92,
> showing a disconnect, then a reconnect, thus causing additional backbone
> traffic. 
> 

Firstly, adding IP addresses is optional. It works perfectly well without.

Secondly, I frequently have a few 100 users on my node and the extra 
overhead of adding the IP address to just A records is small compared to 
the overall traffic on the links. The only time it is actually obvious 
is when one restarts. It would be interesting to compare the amount of 
PC92A traffic versus spots.

Thirdly, having been through one of these abuse storms (actually this 
last one is really only a small line squall) several times in the past, 
  spots are really not the only problem. Announces are usually a weapon 
of choice, as well as injected PC41s.

Getting a different IP address is quite common even on non-dialup lines. 
It isn't the actual IP address that matters, it is when you see that 
call on an IP address from a different ISP (maybe at the same time, on 
some other node) that it becomes interesting.

Not knowing the IP addresses that people that log on with has been a 
perennial complaint for years. This has been compounded, I have to say, 
by my attitude that local logs are for local sysops only (unless they 
care to give that privilege away - which they have to do on a call by 
call basis). This was the abuse storm that actually spurred me into 
writing DXSpider in the first place (the first CVS commit was done early 
this morning - exactly 12 years ago).

I agree that users that login, fire and logout don't leave an PC92A and 
I don't (at least at this stage) propose that to be changed - if it 
becomes a sufficiently serious issue in the future then I will have to 
look at it again.  It is just a parameter change in DXSpider.

PC61 seems like a nice idea until one has to deal with 250 odd nodes all 
on different revisions of the code. There will have to be some kind of 
marker to indicate that "PC61 is accepted here" and then one will need 
to downgrade PC61 -> PC11 on the majority of links that won't accept it.

That is not to say it can't or won't be done, it is just that PC92A is 
more easily upgraded than PC[61]1. If I had a free choice I would rather 
restructure it completely as a PC9x sentence which I can guarantee won't 
  escape inappropriately and will, anyway, degrade gracefully in 
software that handles PC9x if that particular sentence is not handled. 
Also I would want to mark Announces and the replacement for PC41 data.

Oh and anyway, PC61 has been used by CLX (for spots) and I imagine a 
spot with an IP address where it expecting something else, being sent to 
CLX will blow it up (as I believe I did a few times in the early days).

But all of this extra marking is required mainly for spot (or other 
data) sucking. By giving each sysop an audit trail of who logged on with 
which IP address and when, an inappropriate spot or whatever can easily 
be tracked down and obviates the need for marking sentences generated at 
"normal" nodes at all.

Dirk G1TLH