(including use cases)
DarkSide50 files into the simulation. Plan is to initially read an event at a time from the file.
Each event is really two read - the header and data block.
Working on python wrapper scripts to handle configuration (write MPI configurations).
Next - EventStore managing and creation of art event processor instance.
art data interface discussion (raw data product) discussed.
Use cases covered: all on one, use of two machines, use of all six machines.
DAQrate - Marc says it should be straightforward to use cet build tools to build this. (1/2 hour of work)
Is it possible to use the messagefacility and fhicl in the event builder?
Designed to compresses fixed size blocks of 32k or 64k words and return the number of bits in the compressed buffer.
Turn into a module? yes. No OpenMP in this code - we will put it into the module. One full event is about 6MB.
We need to define the data product and output format for the compressed data.
We will need to talk about the configuration of the encoder chunk size to be sure it is flexible enough.
(covered most of this in the event builder discussion)
Problem with OpenMP in art. Initially messagelogger facility might be the issue.
Initial work from cards is done.
MasterProductRegistry is one of the major places that has problems with thread safety. It might be true that this registry is only populated during initialization time, and read-only during the multithreaded time.
Pre-creating multiple scheduled (fixed number) was the plan. This means that all modules will be created up front during initialization.
omp tasks are the mechanism we will use.
A module instance will only be executed by one thread.
The ptr facility will not be used during this exercise.