Forums » Parallel I/O initiative »
Data integrity
Added by Luis Kornblueh over 13 years ago
ON the round table discussion we started talking about data integrity. If we start discussing this we need to include vendors to get an idea on the probabilities that each component the data will pass may fail. I fear that the first place is already main memory. That would mean that we need something like ECC Ram for all things done in memory and checksums
at the very beginning of the life of data sets.
I do not know in how far data integrity is assured for MPI communication? This might be another problem hitting.
Replies (1)
RE: Data integrity - Added by Anonymous over 13 years ago
I really can't see our applications taking on this level of data integrity. Yes we can checksum files or data packets to verify that they have not been corrupted but memory integrity should be handled by hardware ECC or such mechanisms as is done in today's supercomputers.
As somebody raised during the ECMWF workshop panel, there will not be a market for systems that suffer from data integrity issues, so this implies an hardware engineering solution.