eSim_DSN: Web-Enabled Distributed Simulation Network

American Institute of Aeronautics and Astronautics1

ESIM_DSNWEB-ENABLED DISTRIBUTED SIMULATION NETWORK

Nazareth S. Bedrossian1, John Novotny2

The Charles Stark Draper Laboratory, Inc.2200 Space Park Dr, Suite 210

Houston, TX 77058

Abstract

In this paper, the eSimDSN approach toachieve distributed simulation capabilityusing the Internet is presented. With thisapproach a complete simulation can beassembled from component subsystems thatrun on different computers. The subsystemsinteract with each other via the Internet Thedistributed simulation uses a hub-and-spoketype network topology. It provides theability to dynamically link simulationsubsystem models to different computers aswell as the ability to assign a particularmodel to each computer. A proof-of-concept demonstrator is also presented. TheeSimDSN demonstrator can be accessed athttp://www.jsc.draper.com/esim which hostsvarious examples of Web enabledsimulations.

Introduction

For aerospace systems with many subsystems andsubcontractors or design teams that aregeographically dispersed, the capability tocollaborate with each has obvious benefits. For one,location becomes irrelevant thus enabling theapplication of the best talent to a particular project.This is particularly relevant when applied to thesimulation component of any aerospace project.Distributed analysis and simulation can be utilized inthe design and verification phases. In this context,distributed analysis refers to the capability to performanalysis/simulations from geographically dispersedlocations with respect to where the software is hostedor maintained. On the other hand, distributedsimulation means that some or all of the simulation

1 Project Manager, The Charles Stark DraperLaboratory, Inc.; [email protected] Student, Department of Electrical Engineering, RiceUniversity

components or elements are not collocated. Theelements can be software, hardware or general causaloperators. In its most abstract definition, it impliesthat a simulation of a complete system at a singlecomputing facility does not exist. This definition fordistributed simulation is related to a recent trend indistributed computing [1], [2], [3] usually referred toas “grid computing”, which distributes the simulationcomputational load among multiple computers.

In this paper, an approach, referred to as eSimDSN, toachieve distributed simulation capability using theInternet is presented. The eSimDSN middlewarecapabilities are described as well as the benefits ofusing such an approach. An example is presented toillustrate eSimDSN features and capabilities.

Motivation

In the design and development of aerospace systems,it is common for each subsystem to have its ownsimulation. Hence, in order to have integratedsimulations at single or multiple sites requiresassembling the complete system at each computingfacility. Essentially this means that duplicate modelsof each subsystem have to be developed, verified,and maintained. It is easy to see how expensive,inefficient, and risky this process is and how it doesnot easily scale up, i.e. addition of new subsystems ormodification of existing ones.

An aerospace system with many of these attributes isthe assembly of the International Space Station (ISS).Until it reaches its final configuration, ISS is anevolving complex system due to the large number ofassembly stages as well as variety of systemscontributed by various International Partners.

To verify the ISS flight software and control systemsfor each assembly stage, simulations of varyingfidelity and integration level are used by a number ofPartners. The usual practice, which requiressubstantial effort and resources, is to assembleintegrated simulations of various subsystems at these


Partner locations. Also at various times, which cannotbe anticipated a priori, changes are made to the flightsoftware or control systems, which must be reflectedin these simulations. Another aspect worthconsidering is the issue of protecting intellectualproperty. This leads to either a reluctance of Partnersto share models/systems or require substantialsafeguards against unauthorized access to them. Inthe case of safeguards, significant resources as wellas time will have to be expended in order to meet therequirements.

Motivated by the previous discussion, a Web basedapproach, eSim, was developed to provide adistributed analysis and simulation infrastructure. Inthe initial phase, a distributed analysis capability wasdeveloped which allows both interactive and non-interactive access to existing tools/simulations usinga Web browser to enter inputs and view results[4],[5]. A similar approach can be found in [6]. Thesecond phase involved developing a DistributedSimulation Network (DSN), eSimDSN.

Distributed Simulation Network (DSN)

The basic concept of the DSN is to connect, using theInternet, multiple computers that serve simulationelements. With this approach a simulation can beassembled from component subsystems that run onthe respective subsystem owner’s equipment. Thesubsystems interact with each other via the Internet,i.e. the Web is the simulation “backplane” or“wiring” system.

As in [4], the DSN uses a Client-Server construct.The user/Client logs on to a Web site that allowsaccess to the DSN. Either a DSN or non-DSNcomputer can in general serve the Web site. From theWeb site, the user can enter or modify simulationparameters, start the simulation remotely, and viewthe results in text, graphical or animation format. Ingeneral, any subsystem simulation can be served onthe DSN with minimal modifications. Further, actualhardware can be used instead of subsystemsimulations. With this approach standard Internetsecurity features are used to only allow authorizedusers to access simulations via the Web

A fundamental benefit of this approach is that itenables collaboration by competitors because itsolves the intellectual property (IP) problem. The IPproblem is resolved since the DSN only providesaccess to the input/output behavior of a subsystem. Ingeneral, it would be very difficult if not impossiblefor a competitor to perform system identification in

order to determine the model or code embedded inthe particular subsystem. Further, the DSN approachreduces design cycle time since results are availableearlier in the design phase. Other benefits can becategorized as reducing risk and/or reducing cost. Inthe following a listing of benefits per category isprovided, where “simulation” is used to denote anyform of resource that is being shared.

Using the DSN architecture, project risk is reduceddue to the following benefits:§ Simulation source code is protected. It provides

secure access to subsystems simulations withoutaccess to source code or executable.

§ Intellectual property rights are protected .Proprietary methods or modeling details are notrevealed.

§ Prevents unauthorized access. Since accountaccess to computational facilities is not required,unauthorized access to files/accounts isprevented.

§ Reduces simulation errors. Since each subsystemis developed and maintained by the originators,errors caused by maintaining multiple versionsof simulations are prevented. The overallsimulation will also reflect the latestmodifications/updates to each subsystem models.

With the DSN architecture, project cost is reduceddue to the following benefits:§ Development, maintenance and distribution

costs are reduced. It eliminates duplicatedevelopment effort and simplifies configurationmanagement since subsystem simulations ormodels are maintained by their originators.

§ Productivity is increased . Productivity isimproved by standardizing simulations, i.e. allusers experience a uniform simulationenvironment and analysis tools as well asenabling geographically dispersed users tocollaborate on a project. As well, since thesubsystem simulations are developed andmaintained by staff that is expert in thatsubsystem, there is no need for an overallconfiguration management function..

§ Allows fractional access to simulations. Withthis approach “metered” use of simulations isenabled. A simulation becomes just like anyother commodity that can be traded in whateverunits are required by the customer.

Finally, another aspect of the DSN is that it acts as aninformation filter. That is, the framework puts theemphasis on subsystem interfaces rather than theircontent. These interfaces provide the minimalinformation set necessary to define a complete


simulation. This focus on interfaces forces thedevelopment of modular system simulations. As aresult, the DSN approach is dynamically scalablewith minimal or no cost when compared to legacyapproaches which require increasing resources inorder to scale-up.

DSN Architecture

In general, for a distributed simulation, the networktopology should not dictate what kind of simulationscan be run. For instance, a program that models amechanical spring accurately also requires the inputto the spring (the position) and a correct place for thespring output (the force) to go. Network connectionsmust be made to the right locations to enable theinput and the output of the spring. One solutionwould be to hardwire the spring such that it connectsto the rest of the simulation on some other computer.Hardwiring in this fashion dictates that the spring canonly be simulated with the computer it is hardwiredto. The DSN solves this problem by providing theability to dynamically link simulation subsystemmodels to different computers. It may also bedesirable to have the computer that is simulating thespring, to simulate some other component/subsystemof the overall simulation. The DSN achieves thisobjective as well.

There are two basic concepts that allow the DSN toachieve the objectives mentioned above. One is theconcept of hardlinks and softlinks. Hardlinks arecomputer connections that are always constant, i.e.they are hardwired. Softlinks are connections thatmay or may not be made depending on thecomputer’s role in the given simulation. In the DSNthere is only one hardlink; the link between the Clientand the DSN Gateway server. All other links aresoftlinks. They depend on the given simulation.

The other concept is that of a hub-and-spoke typenetwork topology. This type of topology is naturallysuited to simulation of aggregate dynamical systems,e.g. Newtonian or Lagrangian systems. The hubrepresents the inertial dynamics of the system towhich the sum of the applied forces/moments isapplied. The spokes represent the appliedforces/moments. For example, general aerospacesystem can usually be modeled with one blockrepresenting the structural dynamics, another blockcontaining the control system, and another block,which may include disturbance models.

With the DSN, the user can arbitrarily choose thecomputer that hosts the hub. The user can also select

what subsystem simulations are executed on thespoke computers. The DSN is implemented using theJava programming language. Since the Client sideApplet contains the listing of computers that are partof the DSN, this list can be arbitrarily modified toincrease or decrease the number of computers that arepart of the network. The DSN architecture is shownin Figure 1. In Figure 1, the elements labeled Snrepresent subsystem simulation models, which aremade available to the DSN via servers residing ondifferent computers. The ‘I’ and ‘O’ elements arewrappers that enable Input and Outputcommunication between the servers and thesubsystem simulations. The dashed lines indicatesoftlinks between the servers and the main hub thatare used as communication channels. The DSNgateway initializes the blocks, receives parameterinputs and sends simulation output back to the applet.

Figure 1: DSN architecture

Each separate computer in the DSN can be viewed asan individual server, having both input and outputconnections to other. When the simulation isinitialized by the Applet, the input and outputconnections of all the computers are set. Each serveralso has access to a simulation model. For instance,the model may be a linear spring. The simulationmodel for each server is set when the appletinitializes the simulation, along with any parametersthat may need to be set. For the spring example, theparameter is the “stiffness” value used in thesimulation model.

The key point is that the architecture of subsystemservers is identical. Each has the capability formultiple input and output connections, and can storesimulation parameters. There are three layers to eachof the servers. These are the input and outputconnections to the other servers, the simulation


model, and the interface between the input and outputto other servers and the simulation model. Thisarchitecture is shown in Figure 2. Each one of thefunctions outlined above is described in more detailin the following:

§ The I/O layer function is to receive the input,transfer it to the interface, and wait for theoutput. The input and output layer specifies thesimulation model, and the input and outputsampling rates of the server.

§ The Simulation Model represents an arbitrarysubsystem simulation. It receives input andproduces an output. Any general system that hasan input and output can be used.

§ The Interface layer function is to provide thecommunication between the input and the outputlayers and the simulation model. For theexample presented in the following section, theJava Native Interface (JNI) is used to interface tothe simulation model.

Figure 2: Subsystem Server architecture

Finally, it should be noted that the DSN architectureis compatible with various operating systems. Hence,it can be used as a general distributed simulationnetwork infrastructure. In the following section, anexample is presented to illustrate the features of theDSN.

DSN Example

In this section, the DSN features are illustrated usinga proof-of-concept demonstrator. The exampledynamical system is a simple harmonically forceddamped oscillator, identical to the one used in [4].The equations of motion for this example are givenby:

)sin(),() ,( WtAxkfxcfxm =++ &&&

The adjustable parameters/models for this exampleare the friction model ) ,( xcf & and friction

coefficient, c, the stiffness model ) ,( xkf and thestiffness coefficient, k , the mass m, the amplitude A,and the frequency of the forcing function W. Twodifferent models, one linear the other nonlinear, wereused for the friction and stiffness. The simulationwas distributed using the following subsystems; anintegrator block that accepts forces and returnsposition and velocity at various sampling rates, astiffness block that accepts position and returns thespring force, a friction block that accepts velocity andreturns the friction force, and a driving sine blockthat generates the sinusoidal forcing function. Aschematic of this system is shown in Figure 3.

Linking together four computers, which can hosteach one of the subsystems, creates the DSN for thisexample. The computer network is shown in Figure4, which also lists the type of equipment used. Thehub can be assigned by the user to any one of the fourcomputers. In this example, the computer labeledSimba serves as the DSN Gateway. The physicallocation of the computers is shown in Figure 5.

To access this demonstrator, the user starts bylogging on to the general eSim project homepagehttp://www.jsc.draper.com/esim/ , and then selectingthe option “Run Distributed Simulation Oscillatorexample ”. The DSN Web page is shown in Figure 6.From this Web page the user can select whichsubsystem executes on a particular computer from thepull-down menus. Further, the sampling rate for eachsubsystem can be selected independently. The onlyrequirement is that they are integer multiples of thehighest sampling frequency, which for this case is100Hz assigned to the hub or integrator subsystem.Once this has been performed, the user would eithermodify the simulation parameters or begin executingthe simulation by selecting the “Start” button. Oncethe simulation has started the results are displayed inthe graph window. The DSN results were comparedagainst a standalone Simulink implementation of thismodel. The results for one set of simulationparameters are shown in Figures 7-8. Figure 7 showsthe complete 30sec simulation results for theposition, velocity, spring force and friction force.Figure 8 shows a 3sec close-up of the results. Fromthese figures it is evident that the simulation resultsare identical.


Conclusions

In this paper, the eSimDSN approach to achievedistributed simulation capability using the Internetwas presented. With this approach, the capability foraerospace system analysts at dispersed geographiclocations to work collaboratively using simulationscan now be achieved rapidly and with minimalexpense. The eSimDSN approach, architecture andcapabilities were described as well as the benefits ofusing such an approach. A proof-of-concept examplewas presented to illustrate its features andcapabilities. A link to the eSim project homepage,http://www.jsc.draper.com/esim/ , which lists publiclyavailable examples for the complete suite of eSimWeb-enabled capabilities, was also provided.

References

[1] L. Mellon, “Use of Cluster Computing inSimulation”, Simulation Technology Magazine,Vol. 1 Issue 2a, March 1999.

[2] M. Waldrop, “Grid Computing”, MITTechnology Review Magazine, Vol. 105, No. 4,May 2002.

[3] Wayne, R., “Peer (to Peer) Pressure: It’s a GoodThing”, Software Development Magazine, Vol.10, No. 4, April 2002.

[4] N. Bedrossian, J. Jang, J. McManis, and J.Tempelton, “eSim: A Software Architecture forWeb-Enabled Simulation”, 2001 InternationalSymposium on Aerospace/Defense SensingSimulation and Controls, Paper No SPIE-2001-4367-25.

[5] N. Bedrossian, J. Jang, J. McManis, and J.Tempelton, “eSim: Software For Web-EnabledDistributed Analysis And Simulation”, 2001AIAA Modeling & Simulation Conference, PaperNo AIAA-2001-4126.

[6] R. Finsterwalder, and W. Waldraff, “UsingJAVA Applets/Servlets For Web-BasedDistributed Flight Simulation”, 2001 AIAAModeling & Simulation Conference, Paper NoAIAA-2001-4129.


Figure 3: eSimDSN Example –Harmonic oscillator simulation model

Figure 4: eSimDSN Example –Harmonic oscillator simulation network model


Figure 5: eSimDSN Example –Harmonic oscillator simulation network computer map


Figure 6: eSimDSN Example – Distributed forced harmonic oscillator simulation Web page


Figure 7: eSimDSN Example – DSN vs Simulink implementation comparison

Figure 8: eSimDSN Example – DSN vs Simulink implementation comparison: Close-up

Documents

eSim_DSN: Web-Enabled Distributed Simulation Network