[Serval-devel] Separating Data Channel and Command Channel

[Miriam P. V. <mir...@gm...>]

Data Channel vs Command Channel. 
================================


In serval_distributed_manager we can see this kind of code. This is
frecuently repeated in connection_operation and many other processes
both in client and server.

handle_cast({send_broadcast_source_to_server,ClientId,MessageData,Timestamp},State) ->
....

handle_cast({send_unicast_source_to_server,_ClientId,DestinationAddress,MessageData,Timestamp},State) 
...

handle_cast({authenticate_plain_request,UserLogin,UserPassword},State) ->
...

handle_cast({join_vlan_request, VlanId}, State) ->
...

handle_cast({create_vlan_request,VlanId,Description,GroupList}, State) ->
..

handle_cast(vlans_list_request, State) ->
...

handle_cast({leave_vlan_request, VlanId}, State) ->
 
[..many others..]



Erlang's daddy says we can not assume that ordering the pattern-matching
will make the first be evaluated before the other ones. So, even if this
computation time is quite short, if you think of the amount of data
messages that serval is proccesing every second (160 packets sec in
internet???), it could be a big amount of useless effort.

One of the most important tips on improving the performance of Serval
server is to sepparate this channels. Data and control. 

>From my point of view, the best way to implement this separation is
having the data channel in a port and the command channel in another
different. So there will be no code to sepparate this messages in the
server,thus deleting a lot of computation. 

I haven't thought about it yet, but with this separation it could be
necessarie to develop new processes in the server for each client. If we
assume we have to implement the "obervers design pattern" i will talk in
another mail, command channel will encrease with big amount of
computation that shall not be mixed with data's one.


What do you think about this?