Notes |
|
(0000229)
|
hugues
|
2011-01-12 11:13
|
|
Reconfiguration is a complex best: if you change parameters for node B, then all nodes connected to B shall be aware of that, and be also reconfigured. This is costly to support and requires application-wide consensus.
Hence, I'm not favorable to support dynamic reconfiguration: we should clearly specify what we can support at a decent cost.
I agree it would be nice to set configuration parameters when starting the application, but once it is started, we should not allow for it, except in error handlers for instance: if connection to B crash with one connection, give the possibility to use another channel, but even in this case, stability of the system is complex to preserve |
|
|
(0000294)
|
user2
|
2011-02-07 16:56
|
|
I think we can close this issue since the driver configuration using ASN.1 would be the reference configuration directive. |
|
|
(0000295)
|
maxime
|
2011-02-07 19:32
|
|
We should keep the discussion open. The points from Jerome have to be investigated.
Reconfiguration at runtime is something we will need because this is something which happens in embedded systems: we loose a communication line, we switch to another.
At the moment we have no possibility to define protocols for the establishment of communication between two nodes, so as soon as one dies, the others die too and the system becomes unrecoverable.
It is likely that we will use the new ASN.1 configuration mechanism to specify some additional properties for a given interface, but that does not solve everything - not the protocol specification. |
|