|Jul 28, 2014|
|7 years 1 month ||SnowpackInterface: station_name and stationID for SnowStations are
already stored when initially reading the snow cover, not during the
|7 years 1 month ||SnowpackInterface: Reducing the amount of memory needed by read
SN_SNOWSOIL_DATA only once. This structure does not change and
consequently does not need to be sent accross the MPI processes.
|7 years 1 month ||SnowpackInterface: writeOutputSNO now operates on the basis of send
and receives only, the MPI gather method has been discarded, due to
the fact, that if we have many processes, the memory demands for the
master node exceed availability.
|7 years 1 month ||SnowpackInterface: SnowpackInterface and SnowpackInterfaceWorker
have been remodelled in order to only read snow cover data through
the master process. The other processes receive the data by MPI send
and receive operations. This lead to restructurin
|7 years 1 month ||MPIControl: adding scatter, send and receive template methods for
vector<T*> types. Only types that have the << and
>> operator implemented can make use of these functions.
Furthermore there is a master_rank() function, which returns the
|Jul 27, 2014|
|7 years 1 month ||AlpineMain: only master reads landuse, dem and pts - then broadcasts
these. DEMObject had a erroneous << and >> operators.
Please update MeteoIO to the latest version.
|7 years 1 month ||AlpineMain: only master needs to have io.ini present in the
filesystem - Config object is broadcasted to all processes
|Jul 26, 2014|
|7 years 1 month ||TerrainRadiation: Fixing 64-bit issue (size_t != unsigned int)
|7 years 1 month ||MPIControl: adding comments
|7 years 1 month ||SnowpackInterface: when using MPI only the master node is
responsible for output. TODO: let the user decide whether only the
master or all nodes are allowed to do file I/O. The second option
could be relevant when using a shared filesystem.
|7 years 1 month ||MPIControl: Adding a gather method for ints and a pseudo gather
template for vector<T*>. It's a pseudo gather, because
underlying send and receives are performed. The usage of MPI_gatherv
was not possible due to string size restrictions and memory de
|Jul 25, 2014|
|7 years 1 month ||SnowpackInterface: getting rid of that zombie code
|7 years 1 month ||SnowpackInterface: It seems that it's enough to call the
runoff->output(Date&) method from the master node only, since
data is supplied solely through runoff->fillHydroTables.
|7 years 1 month ||SnowpackInterface: now runs with MPI and scales very well - however
two issues remain: interdependecy of modules (runoff is called from
within snowpack) and the writing of data, that is distributed on
multiple MPI nodes (writeOutputSNO, writing runof
|Jul 24, 2014|
|7 years 1 month ||SnowpackInterface: Cleanup before inducing MPI
|7 years 1 month ||MeteoObj: Sanitizing for MPI usage
|7 years 1 month ||MeteoObj: There was a misconception as to which date has been
prepared. The non-master MPI nodes simply assume that data is
available for the correct date. Alternatively we could broadcast the
date available in an async fashion.
|7 years 1 month ||MPIControl: Adding a barrier method that calls MPI_Barrier() or does
nothing in case MPI is not active. This can be used to synchronize
anywhere in the code, mostly useful for debugging.
|7 years 1 month ||Cosmetics, reducing output from ViewFactors
|7 years 1 month ||Reverting changes made to ViewFactors class. There was an error when
computing the view factors - further investigation required.