Rebranding distributed Erlang
Wed Apr 28 13:39:53 CEST 2004
On Wed, 28 Apr 2004 11:01:43 +0200 (CEST), Joe Armstrong <joe@REDACTED>
> BTW - what are the upper bounds on a the number of nodes in a
> distributed Erlang system? - If we assume that joins and leaves are
> rare and that nodes are on-line for long periods is it unrealistic to
> have a DE system with thousands/millions of nodes? - what is the
> limiting factor? - what breaks as the system gets large?
As I understand it, you are most of the time supposed to have connections
between all nodes of the Erlang cluster. A node in your Erlang cluster is
typically connected with all other nodes in the cluster. So, I do not not
the limit here, but I assume it is quite low (regarding the usual scale of
a P2P network). You can partition your Erlang cluster network and have
nodes only linked to some of the Erlang cluster nodes, but they are thus
not aware of the nodes they are not linked to and there is no standard way
to route messages from one node to a node I am not linked to. You need to
code something on top of the Erlang protocol to handle routing.
The current Erlang internode protocol is not yet made for P2P
applications. To handle that you need adequate routing algorithms to be
able to create a huge Erlang cluster without the need to connect all nodes
with all the other, as this does not scale well.
I hope i did not misunderstood something in the way the Erlang internode
protocol is running.
But adding P2P support at the level of the Erlang internode communication
could be a good idea, to make Erlang clusters more scalable and to ease
the development of P2P applications.
More information about the erlang-questions