Hardware Reference
In-Depth Information
include methods of using fast, off-node storage as temporary buffers that move
data to disk-based storage, like the Burst Buffer; and a reconceptualization of
storage as a swarm of storage nodes and clients that work in a self-organized
manner to ensure data safety and performance, embodied by Sirocco. These
technologies can enable legacy applications to attain high-bandwidth peri-
odic checkpointing to the le system with power use proportional to today's
systems.
Other techniques that emphasize client behavior are also available that
can reduce overall demands on the I/O system. These methods reduce the
frequency of checkpointing, allowing longer running checkpoints to take pro-
portionally similar amounts of application time to complete. By making jobs
more reliable, and guarding against the most common types of faults, defensive
I/O load on the system can be dramatically reduced.
These solutions, applied individually and together, significantly increase
the capabilities of limited disk-based I/O systems. Bringing these technologies
to effective use will allow HPC platforms to generate impactful research results
for well into the exascale era.
Bibliography
[1] The Green500 list.
http://green500.org/
.
[2] Intrepid/Challenger/Eureka File Systems.
https://www.alcf.anl.gov/
user-guides/intrepid-challenger-eureka-file-systems
.
[3] TOP500 Supercomputer Sites.
http://www.top500.org/
.
[4] Pawan Agnihotri, Vijay K. Agarwala, Jeffrey J. Nucciarone, Kevin M.
Morooney, and Chita Das. The Penn State Computing Condominium
Scheduling System. In Proceedings of the 1998 ACM/IEEE Conference
on Supercomputing (CDROM), Supercomputing '98, pages 1{23, Wash-
ington, DC, USA, 1998. IEEE Computer Society.
[5] James Ang, Doug Doerfler, Sudip Dosanjh, Scott Hemmert, Ken Koch,
John Morrison, and Manuel Vigil. The Alliance for Computing at the
Extreme Scale. In Proceedings of the Cray Users Group 2010, 2010.
[6] Blaise Barney. Using the Dawn BG/P System.
https://computing.
llnl.gov/tutorials/bgp/
.
[7] Brian Behlendorf. Sequoia's 55PB Lustre+ZFS Filesystem. Lus-
tre User Group 2012,
http://www.opensfs.org/wp-content/uploads/
2011/11/LUG12_ZFS_Lustre_for_Sequoia.pdf
, April 2012.