[GRIDRPC-WG] [saga-core-wg] Grid RPC data management & SAGA

Andre Merzky andre at merzky.net
Thu Sep 3 14:50:36 CDT 2009


Hi Gaël, 

interesting points!  I'd say that the change in semantics
warrant a separate rpc_data class.

But I ahve a question, to help me understand: so far I had
the impression that rpc calls would, in the new model, also
accept a mixture of the old style parameters and the new
style data handles - is that the case?

If yes, then we would need to change the parameters for the
rpc.call() method in a way that both are accepted.  If that
is not the case, and you can cleanly distingish both use
cases, I don't see a problem with a second rpc.call() method
which would overload the original one with a different
signature.

My $0.02, Andre.


Quoting [Ga?l Le Mahec] (Sep 02 2009):
> 
> Hello,
> 
> 	As we propose at the OGF'25 (Catania), we worked on the Grid RPC 
> 	data  management API integration into the SAGA API (see joined documents).  
> We tried to adopt the SAGA Look-&-Feel and we propose two different  
> ways for the integration :
> 
> - Extending the original saga::rpc::parameter class and adding the  
> missing classes/enumerations for a proper integration.
> - Adding a saga::rpc::rpc_data class to the API. This class extends  
> the main saga::rpc::rpc overloading the call method.
> 
> These two different approaches have advantages and disadvantages.
> Using a unique saga::rpc::rpc class implies to modify the original  
> class to respect the SAGA Look-&-Feel. For example, the attributes of  
> the orginal saga::rpc::parameter class do not make sense for the sub- 
> class saga::rpc::parameter_data. Moreover, the original  
> saga::rpc::parameter inherits from the saga::buffer class, that should  
> not be used for the parameter_data class (change in the semantic).
> 
> The second approach is more easily implementable but the API then  
> contains two different methods to call a service.
> 
> Maybe there is a better way to extend the current SAGA RPC avoiding to  
> change the parameter class and the two different methods. Please feel  
> free to send any comments about these propositions.
> 
> Best regards,
> 
> G. Le Mahec for the GRAAL team.
-- 
Nothing is ever easy.


More information about the gridrpc-wg mailing list