Hardware Reference
In-Depth Information
16.5 Conclusion
Future development efforts for HDF5 include the ever-present press to
improve performance, for both parallel and sequential applications. Adding
capabilities to improve concurrent multi-process access to HDF5 files, partic-
ularly in a single-writer/multiple-reader scenario, is underway as well, which
will enable using HDF5 in new data- and event-monitoring scenarios. The ex-
plosion in the number of compute cores available to a single application has
increased pressure to make the HDF5 library multi-threaded and to support
asynchronous I/O, both of which are under development. Efforts to improve
fault tolerance of HDF5 files will enable recovery of files open for modification
when an application crashes.
HDF5 has been under development for over 18 years and is expected to
continue to grow and evolve for many more to come. Major application com-
munities, government agencies, and large corporations rely on HDF5 for stor-
ing mission-critical data that must remain accessible for decades. As new
I/O and computing paradigms emerge, such as using flash-memory for stor-
age, shifting from synchronous to asynchronous I/O operations, and building
fault-tolerant file systems for exascale computing, HDF5 continues to adapt
to the ever-changing landscape in high performance computing.
16.6 Additional Resources
Further information about HDF5 can be found on its home page:
http:
//www.hdfgroup.org/HDF5/
. The HDF5 home page contains extensive doc-
umentation about building and using the HDF5 library and tools, example
programs, tutorials, pre-built binaries, the source code repository, links to
hundreds of packages that use HDF5, and information about how to access
the HDF5 help desk and join the mailing list.
Bibliography
[1] A. Adelmann et al. The OPAL (Object Oriented Parallel Accelerator
Library) Framework. Technical Report PSI-PR-08-02, Paul Scherrer In-
stitut, 2008{2010.
