JLab Software Meeting
September 22, 2005
David Lawrence



Attending: David Lawrence, Elliott Wolin, Steve Wood, Rob Feuerbach
Sergey Boiarnov, Vardan Gyurjyan, Ole Hansen

DANA Analysis Framework
-------------------------

My slides can be found at:

http://www.jlab.org/~davidl/HallD/Presentations/2005_09_22_DANA/


David Lawrence gave an overview of the DANA analysis framework
currently being developed for Hall-D. The framework is an object-
oriented software package written in C++. It provides for
reconstruction software to be written in a modular way where the
framework brokers the communication between modules (factories).

DANA follows a "factory" model in which factory objects are
written that produce data objects of a certain type. C++ templates
are used to identify the factory who produces data objects of the
desired type. Objects are created on demand and then "recycled"
during the event (i.e. pointers and not objects are passed) so
CPU usage is focussed only where it is needed. Factories can
be "tagged" so that more than one factory can be used to produce
the same type of objects. The object pointers are passed in STL
vectors as const pointers.

The framework is designed to process events in multiple threads.
Early testing on SMP machines shows the processing rate scales
approximately with the number of threads (until the number of
threads exceeds the number of CPUs).

Other features such as data object identifiers, configuration
parameters, and event sources are also dicussed.

Discussion
-----------
Most of the discussion was made on clarifying how certain
features are implemented in DANA and how to accomplish
some specific tasks. For example, if one wanted to
re-analyze the same event multiple times, but with some
parameters adjusted between iterations. While I'm not
sure I made the answer to this clear at the meeting, it
can be done with the existing framework like this:

DFactory_base *fac = loop->GetFactory("PID");
DFactory_PID *pidfac = dynamic_cast<DFactory_PID*>(fac);
if(pidfac){
	vector<vector<PID*> > all_pids;  // vector to hold all results
	
	// Loop over all values of param we wish to try
	for(int i=0; i<params.size() ; i++){
		// Call special method of the PID factory to generate
		// the objects using the specified parameter(s)
		vector<PID*> pids = pidfac->TryWithParams(params[i]);
		
		// Remember the results of this for later comparison
		all_pids.push_back(pids);
	}
	
	// .... Compare results ....

	// Delete objects created by TryWithParams() method
	for(int i=0; i<all_pids.size(); i++){
		vector<PID*> pids = all_pids[i];
		for(int j=0; j<pids.size(); j+) delete pids[j];
	}
}



Next Meeting
--------------
The next meeting will take place in approximately 2 weeks on a
date TBD. It was suggested that Hall-B give a presentation to
complete the series.