Hardware Reference
In-Depth Information
computing, data analysis, and other post data acquisition services to support
the experimental facilities.
Recently, formal user requirement reviews [6] have identified a broad range
of storage needs from a ratio of N compute notes to one file (or shared file)
I/O demanding a multi-terabyte-sized file to a ratio of N compute node to
N files (or file-per-process) I/O demanding millions of files per computational
job. Users are interested in checkpointing to improve application resiliency
for a more complex and high-scale computing and storage environment, and
to enable running what would be a very long job in many small jobs using
checkpoints to continue from a previous job.
In preparation for helping science projects ready their applications for
Exascale environments, NERSC user services compiled a chart of the type
and quantity of codes running on NERSC systems (see Figure 2.4. The chart
shows the broad range of codes that have diverse I/O workload needs as well.
FIGURE 2.4: A pie chart of science applications for 2012 that were run at
NERSC. [Image courtesy of Katie Antypas (NERSC).]
