WORKFLOW-BASED DISTRIBUTED ENVIRONMENT

FOR LEGACY SIMULATION APPLICATIONS

Mirko Sonntag

, Sven Hotta

, Dimka Karastoyanova

, David Molnar

and Siegfried Schmauder

Institute of Architecture of Application Systems, University of Stuttgart

Universitaetsstrasse 38, 70569 Stuttgart, Germany

{sonntag, hotta, karastoyanova}@iaas.uni-stuttgart.de

Institute of Materials Testing, Materials Science and Strength of Materials, University of Stuttgart

Pfaffenwaldring 32, 70569 Stuttgart, Germany

Keywords: Simulation workflows, Distributed simulations, BPEL, Web services, Monte-Carlo.

Abstract: Computer simulations play an increasingly important role to explain or predict phenomena of the real world.

We recognized during our work with scientific institutes that many simulation programs can be considered

legacy applications with low software ergonomics, usability, and hardware support. Often there is no GUI

and tedious manual tasks have to be conducted. We are convinced that the information technology and

software engineering concepts can help to improve this situation to a great extent. In this poster presentation

we therefore propose a concept of a simulation environment for legacy scientific applications. Core of the

concept are simulation workflows that enable a distributed execution of former monolithic programs and a

resource manager that steers server work load and handles data. As proof of concept we implemented a

Monte-Carlo simulation of precipitations in copper-alloyed iron and tested it with real data.

1 INTRODUCTION

The importance of computer simulations increases

steadily. Due to achievements in information

technology it is possible to use more and more

complex and hence realistic simulation models. But

there is still potential to improve existing solutions.

In collaborations with scientific institutes in the

scope of our research project we perceived that

many simulation applications are based on legacy

software that was developed over years and is still in

development process.

Many authors contributed to the software and

may already have left the institute or organization.

Usually, there is no time, money or knowledge to re-

implement the tools in a modern programming

language. The software is simply enhanced with new

features. We experienced that there are many

monolithic simulation tools in use that do not benefit

from multi-core CPUs and distributed computing.

The programs often cannot deal with parallel

invocations, e.g. because they do not organize the

result files appropriately. The manual tasks that

scientists have to carry out to achieve their results

are another problem. For instance, they have to

create input files, copy result files between the

participating applications (for calculations,

visualizations, analysis, etc.), they have to start these

applications, or merge results. These and other

problems leave room for improvements of existing

scientific simulation applications.

In this poster presentation we show a concept for

a distributed simulation environment based on the

workflow technology. In the last 5 to 10 years much

research has been done to apply workflows for

scientific applications (Deelman et al., 2004; Barga

et al., 2008; Goerlach et al., 2011; Sonntag et al.,

2010). We are convinced that workflows possess

great potential to improve the tool support for many

scientists. These are the tools scientists work with

every day. There is a need to automate and optimize

simulation processes, to exploit recent developments

in hard- and software, and to improve user-

friendliness and flexibility of simulation

applications. This can be done if modern IT and

software engineering technologies and techniques

are used. At the same time, scientists do not want to

re-write existing code or re-implement applications

for simulation tasks. The proposed concept

addresses these and other requirements.

Sonntag M., Hotta S., Karastoyanova D., Molnar D. and Schmauder S..

WORKFLOW-BASED DISTRIBUTED ENVIRONMENT FOR LEGACY SIMULATION APPLICATIONS.

DOI: 10.5220/0003444400910094

In Proceedings of the 6th International Conference on Software and Database Technologies (ICSOFT-2011), pages 91-94

ISBN: 978-989-8425-76-8

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

The paper is structured as follows. Section 2

presents the use case for the simulation of solids.

Section 3 describes the infrastructure for the

distributed simulation applied to this use case.

Section 4 closes the paper with a conclusion.

2 USE CASE: SIMULATION OF

SOLID BODIES

The macroscopic mechanical properties of metals

strongly depend on their underlying atomistic

structures. Copper-alloyed α-iron, e. g., changes its

material behaviour during the ageing process,

especially when operated at higher temperatures of

above 300°C. In that case, precipitates form within

the iron matrix, yielding to precipitation

strengthening of the material (Kizler et al., 2000)

followed by a decrease of the material strength as

the precipitates grow further in time. This complex

behaviour is modelled with the help of a Kinetic

Monte-Carlo (KMC) approach (Schmauder and

Binkele, 2002) based on a code developed by

Soisson et al. (1996).

Figure 1: Body-centered cubic crystal lattice with a

vacancy-site and neighbour atoms for one possible atom

configuration.

As starting configuration, a number of copper

atoms are placed into a fixed body-centered cubic

iron lattice by exchanging an iron with a copper

atom. A vacancy within the simulation box allows

the movement of an atom by site exchange between

the vacancy and a neighbouring atom (Fe, Cu) (Fig.

1). In every Monte Carlo (MC) step, the jump

frequencies for each of the neighbours of the

vacancy are calculated. By applying a rejection-free

residence time algorithm, one of these possibilities is

selected as the new position of the vacancy. In the

long run, a series of vacancy jumps during the

simulation yields the formation of precipitates with

mean radii above 1 nm (Fig. 2).

Figure 2: KMC simulation of precipitation. Iron atoms are

transparent for better visualization. Start configuration

with randomly distributed copper atoms (A). Precipitates

form during the thermal ageing process (B) (Molnar et al.,

2010).

At desired MC time steps the atom configuration

(i.e. a snapshot) is analyzed yielding particle size

distributions, mean radii of particles and the degree

of supersaturation of the remaining solid solution.

The existing simulation application opal

(Binkele and Schmauder, 2003) consists of five

monolithic Fortran programs where the individual

programs require manually created input data and

are started manually. Two of them build up the

starting configuration and calculate the input values

(e.g. the energies). In addition to iron and copper,

two more atom species can be incorporated. The

precipitation process is simulated by the main opal

application. After specified time intervals, snapshots

are generated. The analysis of these snapshots is

performed by the last two programs which identify

the precipitations within the matrix and determine

size distributions and mean radii. Finally, the results

are visualized, e.g. with MatLab or Gnuplot. All in

all, the overall simulation can be subdivided into

four phases: preparation, simulation, analysis and

visualization.

3 WORKFLOW-BASED

DISTRIBUTED SIMULATIONS

The workflow technology (Leymann and Roller,

2000) can improve the setup and execution of

scientific simulations. With the help of workflows

scientists can adapt the simulation logic without

touching the simulation code (e.g. use another

visualization program). Fault handling features of

workflows languages increase the robustness of

simulation applications, especially if the simulation

functions are implemented with a language that

offers only restricted fault handling mechanisms.

The monitoring component of workflow engines

enables users to observe the progress of their

workflows, i.e. their simulations. BPEL and other

workflow languages enable the invocation of

ICSOFT 2011 - 6th International Conference on Software and Data Technologies

programs in a network using Web services.

Distributed applications can thus easily be

implemented by orchestrating functions located on

different machines. Further, different application

types can be integrated and manual steps can be

automated, e.g. the invocation of a visualization tool

after a simulation run is finished.

The conceptual architecture for a distributed

simulation environment (Fig. 3) shows the

components and their dependencies needed for an

application of workflows in the scientific domain.

Figure 3: Architecture of the simulation environment.

The simulation applications are the domain

specific part of the software. A thin wrapper

publishes the programs in a network so that they can

be invoked and accessed remotely. For the use case,

we extended the interfaces of the five Fortran

applications to allow their invocation with

parameters. Furthermore, we created a WS wrapper

for them. Since there is no WS toolkit for Fortran

programs, we built a Java wrapper that invokes the

Fortran executables.

The resource manager (RM) is a simulation

application registry, data storage and supervisor of

the simulation infrastructure. Participating servers

and simulation applications are registered at the RM

prior to or during the execution of simulations. The

RM can even install software on demand on an idle

server with the help of provisioning techniques as

used in Grid or Cloud computing (Foster and

Kesselman, 2004). Clients that want to use a

simulation application have to ask the RM for

permission. Permission is granted with the help of

service tickets that warrant exclusive usage right.

This is comparable to advanced reservation

techniques in Grids (Foster and Kesselman, 2004).

The goal of the RM is to ensure load balancing.

Furthermore, it checks the availability of simulation

servers to detect network partitions. Clients get

informed so that they can react accordingly (e.g. by

requesting the service again). Finally, the resource

manager can be used as storage for simulation data

(e.g. simulation parameters, configuration, results).

Each simulation run gets its own simulation context

where data can be saved, organized and deleted. The

context correlates data items that belong to the same

simulation run. This improves reproduction of

simulations because the configurations and input

data as well as result data are assembled and can be

observed by scientists at a glance. The data items do

not have to be searched for and correlated later on,

which may not be possible at all since simulation

applications may not have a sophisticated storage

mechanism (e.g. legacy simulation applications

might overwrite simulation data of former runs).

The workflow engine is responsible for executing

the simulation workflows that implement the logic

of the simulations. We implemented the use case

with two BPEL workflows. The main workflow

acquires a service for the MC simulation, executes

the simulation, collects the results and starts the

post-processing workflow for each intermediate

result. The post-processing workflow acquires and

invokes the services for calculating the

precipitations and their radii. If post-processing is

done, the main workflow invokes Gnuplot to

visualize the results.

The simulation manager (SM) offers domain-

specific functions to configure and start simulations

and to access simulation context data. It makes use

of the resource manager as storage facility. In multi-

scale or multi-physics simulations a single

simulation can consist of multiple workflows. The

SM can be used to manage these different

participating workflows.

The simulation client is a GUI used by scientists

to interact with the simulation environment. It

provides the domain-specific functions of the SM

and the general functions of the RM to the scientists.

For the use case we have additionally implemented

functions to store input data as profile for other

simulation runs, a file explorer that shows the

context of simulation runs, and a 3D visualization of

the intermediate lattice snapshots (Fig. 4). The latter

allows observing the convergence of the simulation

results during run time.

For monitoring purposes we have implemented a

view that shows all service ticket requests that are

sent to the resource manager and a view that lists all

issued service tickets. This information enables

scientists to observe the status of the simulation

environment and existing problems (e.g. a long list

of open service requests may indicate too few

available instances of a simulation service in the

system).

ServerWorkstation

Server

Resource Manager

Simulation

Application

SimulationManager

Simulation

Application

Simulation

Application

WorkflowEngine

Legend

Server/

Work s ta ti on

Component

Invocation

Simulation

Client

WORKFLOW-BASED DISTRIBUTED ENVIRONMENT FOR LEGACY SIMULATION APPLICATIONS

Figure 4: 3D preview of the simulation snapshots (Hotta,

2010).

4 CONCLUSIONS

In this paper we have presented a concept for an

infrastructure that enables the distributed execution

of scientific simulations. Core of the infrastructure

are simulation workflows that reflect the logic of

simulations and a resource manager that controls the

work distribution on the participating machines and

that works as storage for simulation data. The

infrastructure addresses main requirements of

scientists on a simulation environment and hence

can improve the tool support for scientific

simulations, e.g. to automate manual tasks, to enable

distributed execution of legacy software.

As a proof of concept we implemented an MC

simulation of solid bodies based on BPEL and tested

it with realistic data. The software improves the

simulation process and the former application to a

great extent. The scientists now have a GUI to start

and monitor their simulations, some of the manual

steps could be automated (e.g. start of post-

processing and visualization of results), and multiple

CPU cores and distributed computing can be

exploited. We are convinced that our consideration

can help scientists with their every day work.

ACKNOWLEDGEMENTS

The authors would like to thank the German

Research Foundation (DFG) for financial support of

the project within the Cluster of Excellence in

Simulation Technology (EXC 310/1) at the

University of Stuttgart. We thank Peter Binkele who

contributed the MC simulation code opal to our

work.

REFERENCES

Barga, R., Jackson, J., Araujo, N. et al., 2008. The Trident

Scientific Workflow Workbench. In Proc. of the IEEE

International Conference on eScience.

Binkele, P., Schmauder, S., 2003. An atomistic Monte

Carlo simulation for precipitation in a binary system.

In International Journal for Materials Research, 94,

pp. 1-6.

Deelman E., Blythe, J., Gil, Y. et al., 2004. Pegasus:

Mapping Scientific Workflows Onto The Grid. In

Proceedings of the 2

European AcrossGrids

Conference, pp. 11-20, Springer-Verlag.

Foster, I., Kesselman, C., 2004. The Grid 2: Blueprint for

a New Computing Infrastructure. Morgan Kaufmann,

edition.

Goerlach, K., Sonntag, M., Karastoyanova, D. et al., 2011.

Conventional Workflow Technology for Scientific

Simulation. In: Yang, X., Wang, L., Jie, W., 2011.

Guide to e-science. Springer-Verlag.

Hotta, S., 2010. Ausführung von Festkörpersimulationen

auf Basis der Workflow Technologie. Diploma Thesis

No. 3029, University of Stuttgart.

Kizler, P., Uhlmann, D., Schmauder, S., 2000. Linking

Nanoscale and Macroscale: Calculation of the Change

in Crack Growth Resistance of Steels with Different

States of Cu Precipitation Using a Modification of

Stress-strain Curves Owing to Dislocation Theory. In

Nuclear Engineering and Design, 196, pp. 175-183.

Leymann, F., Roller, D., 2000. Production Workflow –

Concepts and Techniques. Prentice Hall.

Molnar, D., Binkele, P., Hocker, S., Schmauder, S., 2010.

Multiscale Modelling of Nano Tensile Tests for

Different Cu-precipitation States in α-Fe. In:

Proceedings of the 5

International Conference on

Multiscale Materials Modelling, pp. 235-239,

Fraunhofer Verlag.

Schmauder, S., Binkele, P., 2002. Atomistic Computer

Simulation of the Formation of Cu-Precipitates in

Steels. Computational Materials Science, 24,

pp. 42-53.

Soisson, F., Barbu, A., Martin, G., 1996. Monte-Carlo

Simulations of Copper Precipitates in Dilute Iron-

Copper Alloys During Thermal Ageing and Under

Electron Irradiation. In Acta Materialia, 44,

pp. 3789-3800.

Sonntag, M., Karastoyanova, D., Deelman, E., 2010.

Bridging the Gap Between Business and Scientific

Workflows. In 6

IEEE International Conference on

e-Science.

ICSOFT 2011 - 6th International Conference on Software and Data Technologies