lc2-hdf5-110

Investigate hdf5 1.10 features like SWMR and virtual dataset for LCLS II

Setup

I made a conda environment with hdf5 1.10, h5py 2.7, and python 3.5

The makefile sets PREFIX to the root of this conda environment.

make will compile the C++ code, lib, and make a soft link to the pyhon driver

Running

To run, first do

bin/ana_daq_driver -h

Some options we can set from the command line, but most are controlled through the yaml config file: https://github.com/slaclab/lc2-hdf5-110/blob/master/config.yaml

You run ana_daq_driver and it

• creates root_dir/run_dir
• use --force to overwrite, it always neesds a new directory
• creates root_dir/run_dir/hdf5 - sets lfs striping on it per config file
• creates root_dir/run_dir/logs - and a /results /pids subdirs.
• bin/daq_writer - many of these will run from different hosts. They will write into the /hdf5 in SWMR mode (effectively MWMR since we will have many running in parallel, creating separate h5 files, one for each daq_writer)
• bin/daq_master - one of these should run, it will read all the daq_writer files and make a master file with the virtual dataset
• bin/ana_reader_master - many of these will run, they will all read the master file to work with the virtual view
• bin/ana_reader_stream - we can run these too, they will read the daq_writer streams directly
• all the C++ programs will write their pid's in the pids dir. To clean up, the driver has a --kill command.

daq_writer

The schema will be

/small/00000/data
/small/00000/fiducials
/small/00000/milli

/small/00001/data
...
/vlen/00000/blob
/vlen/00000/blobstart
/vlen/00000/blobcount
/vlen/00000/fiducials
/vlen/00000/milli
...
/detector/00000/data
...

We will write many small groups, and each will have three datases. fiducials will just be a counter, and milli will track milliseconds since program start. milli is just for profiling, not merging, fiducials is for merging.