Hi everybody,
I have some technical questions about filters:
1 - Why aren't filters yet supported with parallel I/O? Is it because
a filter is applied to each chunk separated and with parallel I/O is
possible that parts of the same chunk could be read or written by
multiple processes and it would require a lot of overhead to manage
it? The message that the library gives me is "Parallel I/O does not
support filters yet", will the support be extended soon?
2 - Reading about the VOL I understood that the VOL plugin assigned to
a fapl is called just after the public API. If this is correct the
data pipeline, and with it enabled filters, should be done within the
VOL or after the VOL sent the request to something else (e.g., Native
VOL, under VOL, network VOL). I'm wrong? I got suspicious reading this
[1] presentation, in particular slide 104.
I will really appreciate any help you can provide,
Nicola.
[1] http://extremecomputingtraining.anl.gov/files/2013/07/HDF5-Parallel.pdf
ยทยทยท
--
Nicola Cadenelli
Phone (IT) +39 334 6550271
Office (DE) +49 2461 6196 475
Skype: nicolacdnll