Wednesday, 22 March, 16:30 CET (+0100)
Present: David A Schneider (LCLS), Aaron, Andreas, HJB, MK, TSR (until 17:15), PJ
This was an interesting and longer discussion about the use of NeXus, HDF5 and its special features, and the design of a data collection system. The current protype of the LCLS system uses compound types and VLEN dataset. It relies on SWMR to communicate progress of the experiment to data consumers. Since NeXus does not use compound type and VLEN only for strings (and that is painful), they anticipate some problems for efficiency and latency when moving to an all-NeXus way of working. Message brokers were brought into the discussion, that follow a more network-type dataflow with lower latency than what can be achieved through a filesystem. Apart from individual suggestions, like using NXcollection when sticking to compound types for data that end users are not interested in, no final conclusion was reached at 17:15.
Was completed already, recorded here for reference.