[Dxspider-support] multiple logins + real time user info
djk
djk at tobit.co.uk
Sat Jan 2 20:13:52 GMT 2016
Traditionally, until say five or six years ago, a user would connect
once - to one node - and the cluster of nodes would make sure that that
user could not connect to any other node in the same cluster. This is
one of the functions that PC16/17 traffic was used for, and was the
standard method for determining who was on which node. This part of the
cluster protocol (together with PC19/21 for node connections) had many
problems and was the major reason that I came up with PC92. This
sentence removes, once and for all, the major problems caused by
"looping" when one tried to obtain some resilience in the network by
nodes trying to cross connect to different "branches" of the "tree" that
the original AK1A cluster protocol assumes. But it does also introduce
some other things (features if you will) whose utility could be debatable.
One of the side "benefits" that was discussed at the time that PC92 was
introduced, was that users could now also connect to more than one node
at a time. I could (and still can) see some merit in this - in some
limited circumstances - for users in difficult networking situations.
But that was long time ago and internet connectivity has improved well
beyond the point where this facility is, I believe, useful - or indeed -
desirable. Especially as there (now) seem to be several users who
connect to many nodes as a matter of course. This may even be a
function/feature of some client software that they use (I don't know,
not my department).
I have had a number of private complaints about the amount of bandwidth
that running a node (especially a well connected and popular one) takes.
Having "a lot" of users connecting to more than one node simultaneously
creates a lot of additional administrative traffic, which is there to
allow every node to retain a reasonably accurate view of the cluster to,
amongst other things, control the number of connections that a user has.
But even with all this traffic, the "view" that any node has is likely
to be neither complete nor accurate. This is simply because it takes
time for an update to traverse the network and situations could change
during this process. On a node such as GB7DJK with good, stable links,
it will be pretty close - be still probably not accurate. But for all
that networks have improved - there still seem to be nodes scattered
about that can't seem to stay connected stably.
I am working (very slowly) on a newer version of DXSpider and I would
like to discuss the whole area of user "management". I really do wonder
whether - in the modern cluster - with 6000+ concurrent users we
continue to require "real time" user connectivity information in its
present form.
Please discuss.
Happy New Year
Dirk G1TLH
PS: On a DXSpider node the default maximum concurrent connections that a
user can have to a cluster is three. A sysop can enforce any limit they
like on their own node and I know that some sysops set this limit to one.
More information about the Dxspider-support
mailing list