On Feb 25, 2013, at 5:09 PM, Dries Kimpe <[email protected]> wrote:

* Quincey Koziol <[email protected]> [2013-02-25 16:08:47]:

...

I was thinking more about this, and what about the "group collective" I/O operations that we've kicked around a bit? (Where "group collective" is a sub-group of the processes that opened the file)

Or the alternative: a subgroup-open which can quickly generate a handle for a subgroup given the handle for the larger group.

This would be fine, as long as it was fast - I would expect that the subgroup-opened file handle would only exist for a single HDF5 I/O call (read or write). Quincey