[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[netCDF #OZZ-831187]: Issues with parallel netCDF-4

> Full Name: lynton Appel
> Email Address: address@hidden
> Organization: CCFE
> Package Version: 1.4.01
> Operating System: linux
> Hardware:
> Description of problem:
> 
> (1) Please can you provide a clear explanation of the difference
> between collective and independent operations.  Is a "collective"
> write operation a simultaneous write by all processes to a netCDF
> data file (of course in practice processors cannot be guaranteed to
> write precisely at the same time). Does the data written by each
> processor in this situation identical?

Collective writes mean that each processor will wait for all the others to 
catch up before starting the write. Once the write is complete all processors 
go about their business again.

> 
> (2) Is it correct compressed data cannot be written with the parallel
> netCDF4 interface (the facility does not appear to be offered by
> hdf5).

Correct, it is not possible because it is not known in advance how much other 
processors' data will compress.

> 
> (3) When using the serial netCDF4 library please can you arrange for
> configure to insert the following lines into the netcdf.h file:
> 
> #undef MPI_Comm
> #undef MPI_Info
> #undef MPI_COMM_
> #undef MPI_INFO_NULL
> 
> Otherwise it is not possible to access the MPI library.
> 

I will be changing the way netcdf.h is generated, so that these are not defined 
at all. But not for the 4.1 release...

Thanks,

Ed


Ticket Details
===================
Ticket ID: OZZ-831187
Department: Support netCDF
Priority: Normal
Status: Closed