The MEDEA Experiment
Can You Accelerate Simulation-based Learning by
Combining Information Visualization and Interaction Design Principles?
Christopher J. Garasi
1
, Richard R. Drake
1
, John-Mark Collins
2
,
Rafael Picco
2
and Benjamin E. Hankin
2
1
Sandia National Laboratories, Albuquerque, New Mexico, U.S.A.
2
Ideum, Coralles, New Mexico, U.S.A.
Keywords: Scientific Visualization, Interaction Design, Accelerated Learning, High-performance Computing.
Abstract: The intent of the multipurpose display engineering analysis (MEDEA) experiment was to apply the
principles of computer-mediated learning and “play” in the context of high-performance computing (HPC)
modeling analysis. Our approach involved the development of software workflow based on interaction
design principles using a team of graphic artists, experts in graphics- and touch-based displays, computer
programmers, and scientists. The desired outcome was to develop software to overcome perceived HPC
modeling usage and learning barriers common to scientific modeling and visualization. Using multiple
interaction types, a variety of user workflow experiences were captured (novice/learner, analyst, expert)
resulting in a more intuitive and enjoyable experience with a workflow which fosters accelerated learning.
1 INTRODUCTION
High-performance computing (HPC) simulation
codes have been developed to study both simple and
complex physical phenomena such as gravity,
motions of springs, electricity and magnetism,
weather, and fluid motion. Due to their pedigree
these codes invoke an antiquated workflow
paradigm to setup and execute a simulation. A
simulation “input deck” is the ASCII maturation of a
“card deck” once used to program computers (Jones,
2014). An input deck can contain hundreds of lines
of specialized syntax (sometimes cryptic) which a
software input parser then interprets to select the
appropriate physics, define the computational
domain, establish boundary and initial conditions,
initialize materials and their geometry, and specify
variable definitions and output frequency. The input
deck driven simulation workflow can be
summaryzed as both functional and miserable from
an interaction design point of view.
Interaction design entails “designing interactive
products to support the way people communicate
and interact in their everyday and working lives”
(Preece, Sharp, and Rogers, 2015). In the most
successful cases, interactive software is designed
with the user experience in mind (resulting in
engagement and enjoyment). Learnability is a key
element of interaction design as users commonly
dislike spending a lot of time learning how to use the
software.
The typical HPC simulation code workflow
involves the disjointed steps of editing an input
deck, exiting, executing the code (via command
line), and then using an external software package
for results visualization. Novice users experience
multiple barriers to using HPC codes. These barriers
include access to machines with the required
operating systems to run the codes (e.g. Linux), lack
of familiarity with editing software on those
platforms, command line syntax, as well lack of
familiarity with sophisticated visualization software
(e.g. Paraview, Visit, GiD).
Contrast the novice HPC code user experience
with a user installing and running an application on
their smartphone. The level of sophistication of the
smartphone application is not comparable to a HPC
code, however there is a precedent for ease-of-
installation, ease-of-use (or learnability) and
complete functionality contained within the
application. Smartphone applications are not
designed to stovepipe workflow such that the user
has to enter information in one application, then exit,
and then enter another application to run or post-
Garasi C., Drake R., Collins J., Picco R. and Hankin B.
The MEDEA Experiment - Can You Accelerate Simulation-based Learning by Combining Information Visualization and Interaction Design Principles?.
DOI: 10.5220/0006228202990304
In Proceedings of the 12th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2017), pages 299-304
ISBN: 978-989-758-228-8
Copyright
c
2017 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
299
process. In fact, the smartphone application drive for
quality visualization, interaction, speed, and
availability has been established by feedback from
the two-thirds of the American population (Smith,
2015) who own smartphones and a median value of
54% of smartphone users (billions) in emerging or
developing countries (Poushter, 2015).
It is clear that new application interaction and
design practices which are commonly accepted in
social and recreational smartphone application
activities are a paradigm beyond traditional
approaches (Ito, 2009) for HPC modeling. These
new interaction design practices impact what forms
of learning younger generations value and potential-
ly point the way for new workflow practices to be
adopted. Adoption of these new practices can impro-
ve the quality of the workflow for more mature
generations and creates workflow which is already
natural to the younger developing generations.
The intent of the MEDEA experiment was to
establish a seamless workflow between the user and
the HPC code. To the extent possible the input deck
paradigm should be transparent to the user, requiring
only key values of parametric input and a “run
button”. Post-processing of results should be
provided as part of the workflow with elements that
are aesthetically pleasing. Part of the MEDEA
experiment also involved the creation of the team
with the right expertise to make interaction design
and visualization choices to improve the HPC code
user workflow.
Figure 1: MEDEA introductory view. Placing the power in
the hands of the user is the metaphor being invoked. The
user selects from the three options listed.
2 MEDEA DESIGN OBJECTIVES
Prior to the coding of MEDEA, design objectives
were identified which shaped the development of the
application. There was a strong desire to place the
final tool in the hands of the user in such a way that
the disenfranchised (Ito, 2015) who saw nothing but
barriers to HPC code simulation would become
excited and want to explore physical phenomena
using simulation. The user experience had to bypass
both issues with operating system availability (run
on Microsoft Windows) and the hundreds of lines in
the input deck. Simulation execution should entail
simple and clean parameter input and one-touch
simulation execution.
Advancing the ease-of-use (learnability) of the
application was important. Previous efforts to teach
HPC code users have involved days of training.
Training using simplified input decks with variables
defined within the top 25 lines could take 20 minutes
to teach someone how to run the code. The objective
of MEDEA was to determine if the time to learn
how to load a simulation model and enter the
required input could be reduced to minutes, or even
tens of seconds.
Another important element to MEDEA was that
the interface and results should look “cool”.
Spectacle and fun are commonly used in children’s
software to demonstrate style and status, they are
part of the economy of “cool” (Ito, 2009).
Children’s software typically employs “fun” in order
to maintain focus for a sufficient amount of time in
order to solve problems (exploration). Enthusiasm
associated with fun involves sharing and
demonstrating with others. These are inherent
responses we wished to evoke with scientific
simulation visualization. The child may say a view
looks “cool”. To the more mature individual it is
“cool” not only because of aesthetics but because it
provides visualizations that can be used to explain
physical phenomena and why you obtained a certain
level of system performance. An experience is also
“cool” when one can easily learn from it which is
also pleasing. Learning with increased ease is “fun”.
View establishment and simplicity was also a
desired objective. For a given simulation type there
might exist a commonly accepted view of the
results. This type of analysis should be able to
generated easily and should be a natural result from
the simulation. In the case of a new analysis which
might have been published or presented in another
forum, that view should be easily appended to the
views already generated. In either case, the views
should not be cluttered and the user should be
allowed to interact with the resulting analysis in
order to query points, make comparisons, or
simultaneously view multiple variables and how
they relate to one another.
IVAPP 2017 - International Conference on Information Visualization Theory and Applications
300
3 INTERACTION DESIGN
PRINCIPLES
The workflow for MEDEA was selected in order to
foster user activities in the areas of instruction,
manipulation, and exploration (Preece, Sharp, &
Rogers, 2015). The introductory screen for MEDEA
(see Figure 1) allows the user to select from three
options. These options allow the user to 1)
experience views of information vital to simulations,
2) select and run a simulation, or 3) analyze results
from a simulation through pre-defined visualiza-
tions.
Figure 2: MEDEA view option image. The equation of
state surface which relates pressure to density and
temperature is shown. Distinct boundary lines represent
material phase changes incorporated into the material
model. The user can interact with the 3D surface in terms
of orientation as well as querying points on the surface
using the mouse or touch-screen device.
3.1 View Option
High-performance computing models of physical
phenomena require representations of material
properties (equations of state or constitutive
relationships) as a means of specifying physically-
meaningful responses from materials (e.g. density,
phase changes, temperature, pressure). The resulting
quality of a numerical simulation (which can distort
material both geometrically and through multiple
phase changes) is directly related to the accuracy of
the physics description (discretized mathematically)
as well as the accuracy of the material model
representation. The accessibility of material models
to the user community is typically limited to a list of
available material models with terse descriptions of
tabulated numerical bounds. This method of
operating leaves the user in “in the dark” with
respect to the actual details and assembly of the
material model.
MEDEA’s view option allows the user to
directly access and visualize a material model. The
simple act of viewing the material model itself with
ease is a major leap forward for the HPC
community. The experience is expanded by allowing
the user to interact with and explore the material
model using either mouse- or touch-based
manipulation. Details of the tabulated numerical
values can be easily accessed by selecting points on
the three-dimensional surface (see Figure 2).
Examination of the material model surface
shows areas with both smooth and discontinuous
transitions. For this particular example these
transitions are a result of phase changes of the
material. This view immediately fosters a learning
experience including which phase transitions have
been captured, their location, and potential regions
where the table is valid (extrapolation off of the
table is often allowed but not necessarily physically
meaningful). The view section was also planned to
be used to present two-dimensional performance
data generated by an individual or a community
which can be used for informational purposes, or to
be used in comparison with numerical results.
Figure 3: View of MEDEA’s simulate options. Each XML
file has a corresponding PNG file which provides a
summary of the purpose of the simulation.
3.2 Simulate Option
For some user communities, the simple steps
required to access, read, and edit a HPC code input
deck are so unfamiliar that they form a barrier to
utilization. Attempts to modify already existing
input decks so that the user only has to vary
parameter values at the top few lines of the input
deck have equally failed to overcome the distasteful
workflow of editing, exiting, and then invoking
other commands (typically on a Linux command
The MEDEA Experiment - Can You Accelerate Simulation-based Learning by Combining Information Visualization and Interaction Design
Principles?
301
line) to run the code. It was concluded that a simple
set of interactive parameters with no view of the
input deck was desired to overcome the perceived
barrier.
The simulate option begins by providing a user
with a list of available simulation types to run. Each
option represents an input deck which a super-user
has created for the larger community. In conjunction
with the HPC code input deck is a MEDEA XML
file which specifies which variables require input
values (as well as their default values) and the
correct execution command for the HPC code. Users
can identify which simulation template they desire to
run via PNG file images with graphical and text
information detailing the purpose of the simulation
(see Figure 3). Once a template is selected variable
groupings are displayed (with default values
inserted) which can then be altered using the
keyboard or a touchscreen editor (see Figure 4).
Execution of the template is accomplished by the
user selecting the “simulate” bar.
Figure 4: Example of MEDEA’s simulation input option.
The variables and units for this view are established via
keywords in the XML file. A single or multiple pages of
variable grouping can be established. Once completed the
user selects the “simulate” bar to execute the simulation.
3.3 Analyze Option
Visualization of simulation output can occur
immediately after a simulation has been initiated by
the user, or if a user selects a previously completed
simulation. The visualization is created via output
commands listed in the same MEDEA XML file
which contained information as to which variables
should be input and sent to the HPC simulation.
Prescribed visualizations were chosen to be
displayed immediately after the simulation is
executed. MEDEA currently supports 2D & 3D
plotting of X-Y and surface data.
Visualization is an inherent part of the MEDEA
workflow. A novice can begin to study the results of
a simulation as well as obtain views which have
been generated by a super-user (an individual with
greater experience associated with the simulation).
The prescribed views highlight variables of interest
as well as causal relationships between the variables
thus providing a learning experience via
visualization. Experimental results can also be read
into the visualization in order to perform validation
comparisons (see Figure 5).
Figure 5: Visualization of a simulation as prescribed by
output commands in the MEDEA XML file. Experimental
data (circles) has been added to the image and can be
shifted horizontally in time using the mouse or touch-
screen.
Using the flexibility in the XML file, more
advanced users can then create new visualizations
based on the simulation output. Once created, these
new images can be appended to the previous view
descriptions stored in the XML file. Users have
access to both multiple plots per view as well as
multiple views per analysis. Views can be accessed
by clicking or touching the arrows at the bottom of
the screen.
Highly sophisticated views allowing the user to
perform detailed analysis of the simulation
simultaneously with the material view are also
possible (see Figure 6). On the left of the image is
the material phase view surface. Added to the phase
view is the trajectory of the simulation along the
surface (white dotted line). The nearly vertical red
line indicates the corresponding time between the
surface view and the X-Y plots on the right-hand
side. The right-hand side contains a summary view
of more detailed views which can be accessed using
the horizontal arrows. Each view can be time-
correlated with the surface image by touching any of
the summary plots, thereby moving the time-slider
(vertical bar). The size of the right-hand image can
also be expanded by grabbing the right-hand view
and dragging to the left. With simultaneous views of
IVAPP 2017 - International Conference on Information Visualization Theory and Applications
302
Figure 6: Example of highly sophisticated MEDEA analysis comparing the material model (left) with the simulation
trajectory (white line) with multiple time-based performance summaries (right).
material properties, trajectory on the material model
surface, and correlation to time using other output
seen in the right-hand window, the user can begin to
identify how material property variations impact the
global quantities of the simulation. This type of
detailed analysis is readily available to the user and
can be setup by the super-user in the XML file with
relative ease.
4 PORTABILITY &
INTERACTIVITY
MEDEA is an interface to HPC codes which should
provide the already existing simulation community
the opportunity to more effectively run simulations
and analyze their output. From a computing platform
standpoint this can occur using a Windows desktop
or even laptop computer. However, MEDEA was
also designed with increased portability in mind so
that it could also be deployed on a tablet device such
as the Microsoft Surface Pro.
Two portability scenarios were envisioned when
creating MEDEA. First, for a laboratory technologist
or experimentalist to run a simulation of their
experiment immediately prior to execution in order
to prepare for waveform magnitude and timing.
Second, the ability for a user to run a simulation
during a conference or meeting setting in order to
immediately assess performance based an assertion
made by the presenter.
This type of portability for HPC codes and the
ability to have accelerated cycles of learning is
viewed as a future path forward for high-
performance computational simulation. MEDEA
attempts to encapsulate both the best-practices setup
for simulations as well as corresponding
visualizations. This paradigm helps the novice user
obtain solution quickly and serves as a repository for
institutional knowledge on performance evaluation
and analysis. MEDEA was designed to assist both
the computational community as well as members of
the experimental community who want to perform a
quick analysis with minimal additional overhead.
5 CONCLUSIONS
The intent of the MEDEA experiment was to
radically change the workflow for a HPC code user.
Interaction design principles were incorporated with
information visualization techniques in order to
establish a seamless workflow akin to experiences
found with smartphone applications. For HPC codes
which are input deck driven, the MEDEA interface
can be used to run those applications with little
additional overhead. MEDEA’s use of an XML file
to handle input and output from the HPC code
provide a simple interface which generates a
seamless workflow for the user. This seamless and
aesthetically pleasing workflow is a result of
multidisciplinary collaboration between artists,
programmers, and scientists. Novice users can now
The MEDEA Experiment - Can You Accelerate Simulation-based Learning by Combining Information Visualization and Interaction Design
Principles?
303
be introduced to HPC code simulations using the
MEDEA interface and can learn in 10’s of seconds
what would previously have taken hours to learn.
Since visualization is integral to the MEDEA
workflow the user’s time-to-result has been reduced
to minutes, thereby allowing the user to sample more
input conditions and thereby obtain accelerated
learning on parametric sensitivity.
ACKNOWLEDGEMENTS
The authors would like to thank Maddie Minnis
(Ideum) for the creation of the flaming-hand
animation.
REFERENCES
Jones, C., 2014. The Technical and Social History of
Software Engineering, Addison-Wesley. Boston, 1
st
edition.
Preece, J., Sharp, H., and Rogers, Y., 2015. Interaction
Design: Beyond Human-computer Interaction, Wiley.
West Sussex, 4
th
edition.
Smith, A., 2015, U.S. Smartphone Use in 2015, Pew
Research Center (www.pewresearch.org).
Poushter, J., 2016, Smartphone Ownership and Internet
Usage Continues to Climb in Emerging Economies,
Pew Research Center (www.pewglobal.org).
Ito, M., 2009. Engineering Play: A Cultural History of
Children’s Software, The MIT Press. Cambridge, 1
st
edition.
IVAPP 2017 - International Conference on Information Visualization Theory and Applications
304