Hardware Reference
In-Depth Information
TABLE 23.1: Comparison of analysis execution methods.
Insitu In Transit
Post-
Processing
Analysis
Execution
Location
Within
Simulation
Burst Buer
Separate
Application
Data Location
Within
Simu-
Within
Burst
On Parallel File
System
lation
Memory
Buer
Flash
Space
Memory
Data
Reduction
Possible?
YES: Can
limit output to
only analysis
products.
YES: Can limit
data saved to disk
to
NO: All data
saved to disk for
future use.
only
analysis
products.
Interactivity
NO: Analysis
actions must
be pre-scripted
to run within
simulation.
LIMITED: Data
is not perma-
nently resident in
ash and can be
removed to disk.
YES: User has
full control of
what to load
and when to
load data from
disk.
Analysis
Routines
Expected
Fast-running
analysis opera-
tions, statistical
routines,
Longer-running
analysis opera-
tions bounded
by the time un-
til drain to file
system. Statistics
over
All
possible
analysis
and
visualization
routines includ-
ing interactive
exploration of
the rendered
dataset.
image
rendering.
simulation
time.
data for fast access, an opportunity becomes available to perform data reduc-
tions, data analysis, and/or scientific visualization. Analyzing simulation data
in the burst buffer while it is in flight to the parallel file system is known as
\in-transit" analysis. This is in contrast to both traditional post-processing
analysis and emerging in situ analysis. A post-processing workflow requires
the user to reload the requisite data from the parallel file system to perform
the analysis functions desired, independently of the simulation application
itself. In situ analysis is another emerging option to perform analysis that in-
tegrates the analytics routines directly into the simulation, and processes the
data while it is still resident in the simulation's memory space. Each technique
has its own set of benefits and drawbacks, as shown in Table 23.1, which ulti-
mately implies that a combination of these techniques will be employed going
forward.
