HIERARCHICAL TREES OF UNSTEADY SIMULATION DATASETS
Marek Gayer
1
and Pavel Slavik
2
1
Department of Environmental Engineering
Technical University of Crete
University Campus, 73100, Chania
Greece
xgayer@fel.cvut.cz
2
Department of Computer Science and Engineering
Czech Technical University in Prague
Karlovo náměstí 13, 121 35 Prague 2
Czech Republic
slavik@cslab.felk.cvut.cz
ABSTRACT
We have designed and implemented concept of
hierarchical structures consisting of simulation results
stored in unsteady (time-varying) datasets. It allows
incremental, progressive and easy construction and
replaying of various simulation configurations with
interactive visualization of results. The ability of
modification of simulation boundary conditions is
available in every node of the datasets tree. Every
interactively selected path in the tree corresponds to one
modified simulation solution. We have originally
proposed and tested this concept in combustion processes
simulation powered by a simple fluid simulator, but it can
be easily utilized in general simulation applications,
which use unsteady datasets for results storage and
replaying.
Key Words
CFD, real-time and interactive simulation, fluid
simulation, unsteady (time-varying) datasets.
1. Introduction
Real-time simulation with consequent visualization of
various physical and environmental phenomena and
processes is a common goal of many research projects
and applications. Real-time simulation and visualization
offers many significant benefits, such as possibility to
quickly obtain the results, possibility to get a good
overview of the dynamics of the simulated process and
easy manipulation of the simulated model. Also it allows
us to make interactive changes to the simulated process
with immediate visualization response and visualization
of the results in readable and easily understandable form.
To achieve these goals, optimization of simulation based
on simplifying the physical and computational model and
code optimization are being used. Also, the models which
can be effectively solved on existing computer
architectures and can be easily and quickly implemented
are preferred.
1.1 Unsteady datasets
In many complex cases, the above described effort is still
not sufficient to obtain real-time simulation or it would
lead to significant decrease of the computation
preciseness. In such cases, storage of the unsteady (time
varying or time dependent) simulation results in datasets,
which can be in real-time replayed is used in wide range
of simulation applications. Those datasets allow storing of
one or more computed characteristics. The interactivity is
limited; the additional changes to the already computed
simulation setup and simulation configuration are not
available. The replaying part is limited to the strict time
portion based on the dataset stored on the disk. On the
other hand, unsteady datasets offer full interaction in the
visualization part of the simulation (e.g. changing of
visualized characteristic, zooming to any simulated area,
offering various visualization methods including out-of-
core visualization, post-processing and others). The
typical scientific area of this approach is CFD – where we
can easily have insight of spatial and temporal behaviours
of various physical processes in any area of interest [1],
which is important for applications such as combustion
[2]. For a detailed introduction and overview of unsteady
datasets, we refer the reader to [3] and [4].
2. Our effort
We have proposed and investigated a new concept of
storage representation of unsteady datasets. It is based on
hierarchical tree structures. The tree structures consist of
complete unsteady datasets with simulated results (in each
of the tree node). Storing unsteady datasets can be used in
general simulation applications. It allows us use some
interesting features, such is for example possibility to
increase replaying speed by both saving only one of
several simulated frames to dataset (which can result in
orders less disk space requirements) and skipping frames
when replaying dataset. Instead of using unsteady
datasets, we can also use only partial states to accelerate
simulation, such as pre-calculated fluid simulator states,
which we had developed and used in our previous
443-108
303
research [5], but this approach is limited only to
simulation based on fluid simulator and solvers and does
not allow increase of replaying speed and does not allow
saving only selected frames. We had implemented the
concept of unsteady datasets tree to our pulverized coal
combustion application, but it can also used in other
simulation applications.
2.1 The hierarchical datasets concept
In each of the node, the user of the datasets tree can
choose which one of the connected simulation datasets
should be used. This way, the simulation datasets can be
replayed and also created incrementally from any nodes.
The replaying of the simulation results is similar to
replaying of interactive movies. It allows investigating
various simulation configurations and modifications. In
each node, one of the sub-node datasets can be selected. It
adds very important interactive dimension to the common
datasets. We can replay the selected path in the tree from
top to bottom. It results in sequential replaying of
connected stored states of the simulation with selected
modifications of configuration in each of the nodes. After
finishing running a dataset at any node, the simulation can
easily continue without any dataset (the last computed
values from the dataset are used as boundary conditions
for further simulation) with optional availability of storing
the simulated results to datasets.
2.2 Incremental, reusable solution of the
simulation task
This representation also adds easy possibility of
modification the simulation configuration for both the
simulation solver (during solution) and simulation user
(during replaying). Because each configuration uses the
already computed simulation results in ancestral sub-
trees, it results in less disk requirements, which would be
needed for all of the cases, if solved from the scratch. The
structure and the content of the tree can be in any time
extended and modified easily and interactively by simply
storing another dataset. This allows incremental and step-
by-step solution and creating hierarchy of re-playable
results of the simulated problem. This is namely
advantageous for solution of complex tasks. The new
dataset uses as boundary conditions results of state, from
which it was derived (corresponds to the ancestor node).
Each of the nodes has its unique ID of the tree where it
belongs and unique number of nodes in the tree (e.g. 4/15
in Fig. 1 and Fig. 2). This way, the sequences of
connected simulated task modifications are defined in
every node.
2.3 Interaction
The whole tree can be visualized and implemented to
accept interactive user actions and modifications – e.g.
extending the tree with nodes for another part of
simulation, deleting parts of the partial simulation
solution etc. An example of a datasets tree is shown in
Fig. 1. Each node there represents one stored simulation
dataset. Every interactively selected path in the datasets
tree corresponds to one modified simulation solution. The
current selected path of simulation with configurations
modifications is being highlighted, see Fig. 1. In Fig. 2,
the ability of changing the simulation configuration
(boundary conditions) is being demonstrated. In the first
sub-node, the simulation continues with the original inlet
position. In the second sub-node, we have changed the
direction and spread angle of the combustible injection
from the inlet. Simulation continues with changed
parameters with acceleration of visualization being kept
regardless the interactive change of simulation
configuration in the node.
Selected nodes can be made read-only, to for example
disable accidental modification and deleting of the base
tree structure, which is intended to be used by all users.
Figure 1: Hierarchical tree of simulation datasets
Figure 2: Illustration of interactive change of parameters
of simulation in different sub-node datasets
304
2.4 Visualization of simulation data
Our concept allows subsequent real-time visualization of
the datasets. Wide range of general visualization methods
and approaches suitable for unsteady datasets such as
LIC, contour and colour plate visualization [6], volume
rendering [7], flow volumes [8] various types of
streamlines [9] and other techniques [10], [11] can be
used. Pre-computation based visualization for other
techniques such is the motion maps [12] are also
available. In such case, the pre-computed structures can
be placed together with other data to datasets for selected
nodes. This way, our solution could be possibly extended
also for saving whole particle structures for special
visualization methods such as [13] and for various
encoding and data compression techniques such as [14].
For a detailed list of flow visualization possibilities,
which of them most are also applicable to unsteady
simulation datasets, we recommend [15].
Nowadays, real-time visualization of unsteady data is also
maintained by using techniques and approaches regarding
to new programmable abilities of current graphical
accelerators, such as described in [16], [17] and [18]. We
are also currently working on accelerated, real-time b-
spline interpolation based grid visualization of our
simulation results, implemented using vertex and pixel
shaders [19].
2.5 Zooming features and time navigation
We can set any zoom level and change the current view
position to any area of the simulated task under
investigation in our simulation model. Using the unsteady
datasets, we can furthermore arbitrary seek to selected
time position in dataset. We can easily change the
replaying speed (slower, faster) of the simulation or even
back-reverse the combustion process. It is important
because we can easily seek to a particular part of the
simulation as easy as we can set an area of concern in
space. We can control the time resolution and minimum
speed of simulation. For example, when we are creating
an interactive hierarchical unsteady datasets for
simulations based on unstable fluid simulators [20], [21]
we can increase the time step to make the simulation more
precise and at the same time store only selected frames to
be able to replay the simulation at the original simulation
speed without unnecessary frames. This, way the common
drawback of unstable fluid simulation, that forces at the
interactive level to slow computation and interaction due
to too small time level required can be easily avoided.
3. Implementation recommendations
We had implemented our concept to the 2D datasets and
only for those cases, when all of the frames can fit in
main computer memory. But our concept can also be used
for large datasets, even those which would need out-of-
core processing. In our case, we stored the data from the
structured grid and particle system. The general
implementation of the concept depends on the type of the
solved task.
It is important of choosing an optimal storing
representation of the whole structure. Here are many
solutions possible. In our test implementation for
combustion simulation, we use a straightforward
approach, when all the dataset for nodes are stored in
separate binary files.
3.1 File representation and conventions
The binary dataset file consists of a header at the
beginning of file, where fields such as number of total
frames are stored. Also, index of all frames at the end is
present for enabling faster seeking possibility. After the
header, the single frames are successively stored. The
frames consist of arrays of characteristics of structured
grid and all particles, which are needed to restore the
particle system. We use 64-bit file access pointers to be
able to work with files larger than 2 gigabytes.
For each node, we also keep a plain text file with basic
configuration, which contains the initial simulated task
boundary conditions. Applying these boundary
conditions, we prepare the simulation first frame. We can
either continue the whole computation, or we can use the
dataset to replay the stored data.
For the simulation tree nodes stored in our proposed file
organization, we have specified the following file naming
convection. Each file name consists of a number of the
tree and number of the node. In the beginning, we restore
the whole tree by reading the list of files, sorting them and
then recreating the tree from the top to the leaves.
Extending the tree with new nodes is based on saving
other files with appropriate names to the disk and
extending the tree with new dataset. The example of a
name would be test-002-007, where test is the name of the
solved case, 2 is the ID number of the tree and 7 is ID
number of the node.
3.2 Other implementation possibilities
The above described concept is only one possible
solution, not a strict definition. It is also possible use
different ones, which would fulfil the needs of various
ranges of solved tasks. For example, the list of the nodes,
frames and the tree structure and boundary conditions and
configuration could be stored in XML files and binary
datasets of the whole tree nodes could be stored in a one
file instead of many. Any other storage such as a database
based solution is applicable too.
3.3 Unsteady datasets tree applications
Using the above described concept, we can implement
interactive and fast simulation applications. The unsteady
datasets tree has been originally implemented and tested
for our fluid application for combustion processes
305
simulation. But it can be straightforward reused for others
simulation and visualization applications, that either use
or can be extended to use unsteady datasets for storing the
results. This could possibly include for example also
computer graphics field, where tasks such are animation
of liquids and water [22] and gaseous phenomena [23] are
being currently solved.
4. Tests and measurements
We have successfully implemented our concept of
hierarchical datasets to a structured grid based, unstable
fluid simulator for simulation and visualization of
combustion processes. This simulator has been described
in [21]. We have simulated the beginning of pulverized
coal combustion in a 2D approximation of a power plant
boiler. We have compared interactive features of direct
simulation and visualization for both the performance and
interactive features and requirements.
We have measured 2 cases. In the first case, the grid was
set to 20 x 40 cells. In the second case, the grid was set to
50 x 100 cells, and the computation code was more
precise due to a decreased time step.
As a storage drive for a tree of unsteady datasets we used
a 120GB Seagate Barracuda Ultra Ata V. All results and
measurements were performed on a commodity AMD
Athlon 1333Mhz system equipped with 512MB SDRAM
PC 133 memory.
The goal of all of these cases was to provide an
interactive simulation reflecting changes of every 0.05
seconds in a boiler. The SIM case shows the direct
simulation without of using hierarchical datasets. The
UDT is the same case but with interactivity based on
datasets trees. Because of using unstable type of fluid
simulator, the simulation time step had to be set to 0.005s,
which results in more frames being computed. Choosing
less value would lead in the test case to computation
instabilities and errors. The basic features of the task are
summarized in the Table 1.
Features
SIM
UDT
Interactive change of the
simulation
Full
In every tree
node
Simulation time step set 0.005s
0.05s
Total played frames
10000
1000
Space shift and zoom
Yes
Yes
Time zoom features
No
Yes
Table 1: Setup and features comparison of direct
simulation (SIM) and unsteady datasets tree (UDT)
Each of the cases contained about 10000 particles per
animation frame. The grid size has been set to 20 * 40 and
50 * 100 cells. The results are summarized in the Table 2.
In UDT cases, we received acceleration of 24 and 70 of
the total simulation replaying time compared to direct
simulation and immediate visualization. Storing the data
to disk drive consumed virtually no additional time cost
compared to the version without storing the data (with
fast Ultra DMA 5 data transfer mode used).
Store method /
Grid size
SIM /
20*40
UDT /
20*40
SIM /
50*100
UDT /
50*100
Simulation
time 1089s
1103s 4617s 4665s
Write [MB/s]
N/A
0.7
N/A
0.4
Replay time
1062s 49s
4659s
68s
Read [MB/s]
N/A
15.9
N/A
26.4
Drawing speed
[FPS]
9.4 22.5 2.1 14.7
Disk space [GB] N/A
0.78
N/A
1.8
Total
acceleration
N/A
x 24
N/A
x 70
Table 2: Results gained using direct simulation and full
datasets
The amount of data stored in unsteady datasets is in our
case limited to the specific storage drive used. When
running demanding tasks with large cell grids, multiple
characteristics and very long simulation time or frames,
this method would suffer from both capacity and
performance limitations of the drive. Then due to the
speed limit of the drive, a performance bottleneck could
appear when replaying the simulation. This could only be
fixed by storing only selected frames (e.g. only 1 from 10
or even only 1 from 100 and more), but with losing the
precision and resolution of the data stored. But in many
cases, such as in commonly used unsteady fluid
simulators, it is not very important, because the
computation step differs from the simulation step and
saving only selected frames is often fully sufficient.
5. A demonstration example of our results
The following figures display example simulation results
of our pulverized coal combustion application based on
unsteady datasets trees. Our application can arbitrarily
select and switch among a total of about 50 cell and
particle characteristics and statistics in the visualization
phase. Fig. 3 shows a captured frame of real-time
visualization of cell temperatures together with floating
virtual coal particles inside the boiler chamber area. The
results were directly simulated using our simple fluid
simulator [21]. The same results and features (including
switching to all other characteristics and statistics, see
Fig. 4) can be gained using the unsteady datasets tree,
with interactive possibilities of changing the simulation
configuration in each of the node. In such a case, the
system runs in orders faster than the original simulation.
Because of ability of saving only necessary frames in
fluid simulation, the disk space demands are acceptable.
306
In our test it consumed 1.8GB of disk space for fluid
simulator states per 68 seconds animation.
Figure 3: Sample visualization output of our coal
combustion simulation application.
Figure 4: Our application runs using unsteady datasets
tree enabled in orders faster then without them
6. Future work
By implementing optimal existing compression and
encoding methods to unsteady datasets files storage, we
plan decrease both the disk space and traffic requirements
for complex tasks.
We plan testing our system on very large or even 3D
datasets and perform measurements for these cases.
Currently, we are also working on several techniques
based on interpolation, which would considerably
decrease the disk space requirements, while keeping the
precise and smooth visualization of the simulation data’s
stored in unsteady datasets tree structures.
7. Conclusion
We have proposed, fully implemented and tested the
original concept of hierarchical datasets tree structures in
combustion processes simulation and visualization
powered by unstable fluid simulator. Choosing unsteady
dataset helps to avoid the common drawback of slow
simulation of unstable fluid simulators due to small time
steps.
Our concept overcomes the common drawback of the
commonly used unsteady datasets, which results in
loosing of interactivity on the simulation side. Proposed
solution of hierarchical structures consisting of simulation
results datasets over classical datasets allow incremental,
progressive and easy construction and replaying of
various simulation configurations with interactive
visualization of results. The modifications of simulation
boundary conditions are available in every node of the
datasets tree. Every interactively selected path in the
datasets tree corresponds to one modified simulation
solution.
Our results can be easily utilized in general simulation
and modelling applications, which use unsteady datasets
for storage of the results. Users of simulation applications
can now extend the prepared and pre-calculated
simulation results with their own modifications with
possibility to use the results of previous solutions stored
in tree while keeping a high speed of replaying, because
of utilization of unsteady datasets.
8. Acknowledgment
This project has been partially supported by the Ministry
of Education, Youth and Sports of the Czech Republic
under research program No. Y04/98: 212300014
(Research in the area of information technologies and
communications).
References
[1] Bryson, S., and Johan, S. Time management,
simultaneity and time-critical computation in interactive
unsteady visualization environments. In Proceedings of
the 7th conference on Visualization '96 (1996), IEEE
Computer Society Press, pp. 255-ff.
[2] Gillespie, J. The use of CFD in design of combustion
equipment. Progress in Computational Fluid Dynamics 1,
1/2/3 (2001), 80-90.
[3] Cox, M. Large data management for interactive
visualization design. In Proceedings of the SIGGRAPH
'99 System Designs for Visualizing Large-Scale Scientific
Data course notes (1999), ACM Press, pp. 5-29.
307
[4] Bryson, S., Kenwright, D., Cox, M., Ellsworth, D.,
and Haimes, R. Visually exploring gigabyte data sets in
real time. Communications of the ACM 42, 8 (1999), 82-
90.
[5] Gayer, M., and Slavík, P. Pre-calculated fluid
simulator states tree. In Proceedings of 12th IASTED
International Conference on Applied Simulation and
Modelling (2003), IASTED, Anaheim: Acta Press,
pp. 610-615.
[6] Robbins, K. A., and Gorman, M. Fast visualization
methods for comparing dynamics: A case study in
combustion. In Proceedings of the 11th IEEE
Visualization 2000 Conference (VIS 2000) (2000), IEEE
Computer Society.
[7] Crawfis, R., Shen, H.-W., and Max, N. Flow
visualization techniques for CFD using volume rendering.
In Proceedings of the 9th International Symposium on
Flow Visualization (2000).
[8] Becker, B. G., Max, N. L., and Lane, D. A. Unsteady
flow volumes. In Proceedings of the 6th IEEE
Visualization Conference (1995), G.
Nielson and
D. Silver, Eds., IEEE Computer Society Press, pp. 329-
333.
[9] Jobard, B., and Lefer, W. Unsteady flow visualization
by animating evenly-spaced streamlines. In Computer
Graphics Forum (Proceedings of Eurographics 2000)
(2000), vol. 19 of 3, Blackwell Publishers.
[10] Lum, E. B., Ma, K. L., and Clyne, J. Texture
hardware assisted rendering of time-varying volume data.
In Proceedings of the conference on Visualization 2001
(2001), IEEE Computer Society Press, pp. 263-270.
[11] Bauer, D., Peikert, R., Sato, M., and Sick, M. A case
study in selective visualization of unsteady 3D flow. In
Proceedings of the conference on Visualization '02
(2002), IEEE Computer Society, pp. 525-528.
[12] Jobard, B., and Lefer, W. The motion map: efficient
computation of steady flow animations. In Proceedings of
the 8th conference on Visualization '97 (1997), IEEE
Computer Society Press, Los Alamitos, US, pp. 323-328.
[13] Guthe, S., Gumhold, S., and Straßer, W. Interactive
visualization of volumetric vector fields using texture
based particles. Journal of WSCG 10, 3 (2002).
[14] Ma, K., Smith, D., Shih, M., and Shen, H. Efficient
encoding and rendering of time-varying volume data.
ICASE Report No. 98-22 NASA/CR-1998-208424,
Institute for Computer Applications in Science and
Engineering, NASA Langsley Research Center, 1998.
Presented at SIGGRAPH '99.
[15] Hauser, H., Laramee, R. S., and Doleisch, H. State-
of-the-art report 2002 in flow visualization. Tech. Rep.
TR-VRVis-2002-003, VRVis Research Center, Vienna,
Austria, 2002.
[16] Weiskopf, D., Hopf, M., and Ertl, T. Hardware-
accelerated visualization of time-varying 2D and 3D
vector fields by texture advection via programmable per-
pixel operations. In Proceedings of the Vision, Modeling,
and Visualization VMV '01 Conference (2001), T. Ertl,
B. Girod, G. Greiner, H. Niemann, and H.-P. Seidel, Eds.,
pp. 439-446.
[17] Telea, A., and van Wijk, J. J. 3D IBFV: Hardware-
accelerated 3D flow visualization. In Proceedings of the
IEEE Visualization '03 (2003), ACM Press.
[18] Weiskopf, D., and Ertl, T. GPU-based 3D texture
advection for the visualization of unsteady flow fields. In
WSCG Short Communication Papers proceedings (2004),
UNION Agency - Science Press, Plzen, Czech Republic.
[19] Kadlec, P., Gayer, M., and Slavík, P. Visualization
using hardware accelerated spline interpolation. In WSCG
Short Communication Papers proceedings (2004),
UNION Agency - Science Press, Plzen, Czech Republic.
[20] Foster, N., and Metaxas, D. Realistic animation of
liquids. Graphical Models and Image Processing 58, 5
(1996), 471-483.
[21] Gayer, M., Slavík, P., and Hrdlička, F. Interactive
system for pulverized coal combustion visualization with
fluid simulator. In Proceedings of the Visualization,
Imaging, and Image Processing Conference (2002),
IASTED, Anaheim: Acta Press, pp. 288-293.
[22] Enright, D., Marschner, S., and Fedkiw, R.
Animation and rendering of complex water surfaces. In
Proceedings of the 29th annual conference on Computer
graphics and interactive techniques (2002), ACM Press,
pp. 736-744.
[23] Stam, J. Stable fluids. In Proceedings of Computer
Graphics and interactive techniques (1999), vol. 33,
ACM Press, pp. 121-128.
308
