Moved the file open/close back inside the loop in the SWMR reader
code. Without the open/close, the cache never refreshes so we
always see the initial amount.
Made a couple of other minor changes to the SWMR code.
Tested on:
jam
- Added comments to files, functions, etc.
- Some readers changed to keep the file open throughout the
life of the program instead of closing it before sleeping.
This is in the hopes that there will be more opportunities for
cache issues without the close.
- Other minor changes.
Tested on:
jam
Updates the SWMR testing
- All readers/writers now consume and emit their random seeds
for debugging.
- Output uses unbuffered I/O for more readable output.
- The random seeds are now based on a more fine-grained time
measurement to ensure each reader has a different random
seed.
Tested on:
jam
Description:
Adds SWMR support to the v2 b-trees when used as a chunk index. Because each
node keeps track of the total number of records reachable through each of its
children, nodes must be shadowed every time a descendent record is inserted or
removed. This implementation prevents this from happening, however, if the node
has already been shadowed since the last time the v2 b-tree header was flushed.
Also modified SWMR test to include v2 b-trees.
Tested: jam, koala, ostrich (h5committest), durandal
Note: There is a preexisting failure in the ph5diff (-v) test
Bring r22085:22251 from trunk to revise_chunks branch.
Also tackle some testing issues in test/objcopy.c test and clean up
some warnings.
Tested on:
FreeBSD/32 8.2 (loyalty) w/gcc4.6, w/C++ & FORTRAN, in debug mode
FreeBSD/64 8.2 (freedom) w/gcc4.6, w/C++ & FORTRAN, in debug mode
Linux/32 2.6 (jam) w/PGI compilers, w/default API=1.8.x,
w/C++ & FORTRAN, w/threadsafe, in debug mode
Linux/64-amd64 2.6 (koala) w/Intel compilers, w/default API=1.6.x,
w/C++ & FORTRAN, in production mode
Solaris/32 2.10 (linew) w/deprecated symbols disabled, w/C++ & FORTRAN,
w/szip filter, w/threadsafe, in production mode
Linux/PPC 2.6 (ostrich) w/C++ & FORTRAN, w/threadsafe, in debug mode
Linux/64-ia64 2.6 (ember) w/Intel compilers, w/paralle, C++ & FORTRAN,
in production mode
Mac OS X/32 10.7.3 (amazon) in debug mode
Mac OS X/32 10.7.3 (amazon) w/C++ & FORTRAN, w/threadsafe,
in production mode
Mac OS X/32 10.7.3 (amazon) w/parallel, in debug mode
Description:
Adds SWMR capability to v1 b-trees, and the chunk index using v1 b-trees. With
this implementation, flush dependencies are always on when in the cache. This
will allow attritbutes to be used for "checkpointing" data when object header
dependencies are fixed - i.e. if a writer writes data before an attribute in
that dataset's object header, then if a reader sees the updated attribute the
written data is guaranteed to be visible, as long as that dataset's b-tree nodes
are evicted from the reader's cache.
Also adds support for compression with SWMR.
Also fixes earray implementation to not free (reuse) the file space for deleted
chunks and outdated versions of compressed chunks when doing SWMR writes. These
should eventually be added to a timeout list.
Adds testing for these cases.
Tested: durandal
