Hardware Reference
In-Depth Information
the Engineer Research and Development Center of the U.S. Army Corps of
Engineers. At other locations, application scientists install ADIOS themselves
for their own applications. Many of those receive direct guidance from the
ADIOS team to utilize their target system optimally.
17.4 Conclusion
ADIOS provides a scalable I/O library and has successfully improved the
I/O performance of several large-scale applications. In general, it turns out
that scientists are frequently limited by I/O performance, and they limit their
output data accordingly. As ADIOS provides better application I/O perfor-
mance, scientists are now able to write more data and thus gain more scientific
insights during their analysis. With computing systems generating data at an
ever-increasing rate, good application I/O is becoming harder to maintain.
Approaches such as in situ data analysis and data reduction are becoming
more mainstream. The researchers behind ADIOS are working on in situ and
in-transit (or staging) processing capabilities with an easy-to-use API while
maintaining the effectiveness of the ADIOS framework. However, one of the
challenges of staging is that moving data around while the application is run-
ning interferes with the application's own communication, and thus slows the
application down. Therefore, the overhead of staging processing has to be care-
fully identified and analyzed. ADIOS is becoming a staging framework that
provides solutions for low-overhead data transfers. ADIOS's generic I/O API
uses flexible programming that is portable to newer computing systems. It is
also flexible in switching from file-based post-processing to in situ processing.
Bibliography
[1] Janine C Bennett, Hasan Abbasi, Peer-Timo Bremer, Ray Grout, Attila
Gyulassy, Tong Jin, Scott Klasky, Hemanth Kolla, Manish Parashar, Va-
lerio Pascucci, et al. Combining In-Situ and In-Transit Processing to En-
able Extreme-Scale Scientific Analysis. In High Performance Computing,
Networking, Storage and Analysis (SC), 2012 International Conference
for, pages 1{9. IEEE, 2012.
[2] E. Wes Bethel, Hank Childs, and Charles Hansen, editors. High Perfor-
mance Visualization: Enabling Extreme-Scale Scientific Insight. Chap-
man & Hall, CRC Computational Science. CRC Press/Francis{Taylor
