A Topic-Based Data Distribution Management for HLA

Alberto Falcone

and Alfredo Garro

Department of Informatics, Modeling, Electronics and Systems Engineering, University of Calabria,

Via P. Bucci 41/C, Rende, Italy

Keywords:

Distributed Simulation, High Level Architecture (HLA), Data Distribution Management (DDM), Data

Reduction, Topic-Based Publish-Subscribe.

Abstract:

Modeling and Simulation (M&S) represents a fundamental technology for designing and studying complex

systems in various industrial and scientiﬁc domains, when real-world testing is too costly to perform in terms

of safety, time, and other resources. To promote the reusability and interoperability of simulation models al-

lowing them to interoperate without geographic constraints, distributed simulation has been introduced. One

of the most widely adopted standards for distributed simulation is IEEE 1516-2010 - High Level Architecture

(HLA). Among the services provided by HLA, a key one is the Data Distribution Management (DDM) that

allows to reduce the transmission and reception of unnecessary data in order to improve communication effec-

tiveness among simulation models. Although many matching algorithms have been proposed in the literature,

the upcoming HLA 4.0 standard deﬁnes a DDM that still relies on performing matching veriﬁcation by calcu-

lating the overlap between regions using their dimensions. In this paper, a novel topic-based publish-subscribe

messaging system is proposed to improve the performance, reliability, and scalability of DDM services. Ex-

periments show that the proposed topic-based approach achieves better performance than the standard one.

1 INTRODUCTION

In the last years, the complexity of Cyber-Physical

Systems (CPSs) has increased exponentially, mainly

due to the heterogeneity of the involved components

and related interactions that connect the cyber world,

through a network of interconnected computational

resources (e.g., sensors, actuators, and processing

units), to the physical one (Lee, 2008; Falcone et al.,

2020; Falcone et al., 2022). CPSs are characterized

by being highly automated, intelligent, and collabora-

tive. The design and implementation of CPSs repre-

sent a challenging task due to the vast network and

computing resources connected to the physical en-

vironment involving multiple domains such as con-

trols, network protocols, and software engineering.

To capture the structure and behavior of such sys-

tems, researchers rely on Modeling and Simulation

(M&S) techniques (Bouskela et al., 2021; Derler

et al., 2012; Falcone et al., 2014). M&S represents

a pillar technology for designing and studying com-

plex systems in various industrial and scientiﬁc do-

mains, when real-world testing is too costly to per-

https://orcid.org/0000-0002-2660-1432

https://orcid.org/0000-0003-0351-0869

form in terms of safety, time, and other resources. To

promote the reusability and interoperability of simu-

lation models allowing them to interoperate without

geographic constraints, Distributed Simulation (DS)

has been introduced (Fujimoto, 2000). One of the

most widely adopted standards for distributed simu-

lation is IEEE 1516-2010 - High Level Architecture

(IEEE Std. 1516-2010, 2010).

The HLA standard deﬁnes a generic architecture

to support reusability and interoperability across sim-

ulation models. The standard was developed in 1996

under the guidance of the United States Department

of Defense (DoD) Modeling and Simulation Coordi-

nation Ofﬁce (M&S CO). After its deﬁnition, HLA

caught the interest in the industrial and scientiﬁc do-

mains, so it was later moved into an IEEE interna-

tional standard with the ofﬁcial name “IEEE 1516”.

In the HLA terminology, a distributed simulation is

called Federation, which is composed of many HLA

simulation applications called Federates. Federates

interact with each other, in the same Federation, using

the services provided by the Run-Time Infrastructure

(RTI) that represents the communication middleware

that implements the HLA interface speciﬁcations and

rules. The structure and semantics of the data ex-

changed are delineated following the Object Model

186

Falcone, A. and Garro, A.

A Topic-Based Data Distribution Management for HLA.

DOI: 10.5220/0012051500003546

In Proceedings of the 13th International Conference on Simulation and Modeling Methodologies, Technologies and Applications (SIMULTECH 2023), pages 186-193

ISBN: 978-989-758-668-2; ISSN: 2184-2841

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

Template (OMT) speciﬁcations. The RTI provides six

service groups to handle the distributed simulation ex-

ecution: Federation Management, Declaration Man-

agement, Object Management, Ownership Manage-

ment, Time Management, and Data Distribution Man-

agement (DDM).

DDM provides a set of services to minimize the

transmission and reception of unnecessary data and

then maximize communication effectiveness among

Federates, letting them specify the data of interest.

Speciﬁcally, the interest in speciﬁc information is ex-

pressed as bounded portions of a n-dimensional space

of user-deﬁned dimensions. Both consumer and pro-

ducer federates specify the upper and lower bounds

for a speciﬁc portion, thus creating a so-called re-

gion. Federates that produce data deﬁne update re-

gions; whereas, federates that consume data specify

subscription regions.

At the core level, DDM services use an algorithm

that scans all n-dimensional regions to ﬁnd the pairs

of regions (update,subscription) that generate over-

lap. The RTI routes data from producer federates to

consumer ones if and only if there is an overlap, i.e., a

match between their update and subscription regions.

According to the HLA standard, a match must be re-

ported to the RTI exactly once by the DDM services.

This problem is well-known in theoretical computer

science and can be solved using suitable algorithms

with ad-hoc spatial data structures. However, many

DDM implementations tend to rely on less efﬁcient

algorithms that adopt complex spatial data structures

whose manipulation may have a signiﬁcant impact

on computational resource utilization (Marzolla and

D’angelo, 2020).

In this paper, the novel Topic-Based Matching Al-

gorithm (TBMA) is proposed to improve the perfor-

mance, reliability, and scalability of DDM services.

TBMA deﬁnes the concept of topic to handle a high

number of regions, where the match operation is per-

formed by using topics instead of calculating overlaps

between regions’ coordinates.

The paper is structured as follows. Section 2

provides the problem statement along with key def-

initions. Section 3 discusses related work on the

HLA DDM services and existing matching algo-

rithms. Section 4 describes the proposed topic-based

DDM approach. Section 5 presents experiments car-

ried out to evaluate the performance of the proposed

topic-based matching algorithm, where the simulation

results have been compared with the standard one. Fi-

nally, conclusions are discussed in Section 6.

2 PROBLEM STATEMENT

The DDM services are based on the following deﬁni-

tions (IEEE Std. 1516-2010, 2010):

Deﬁnition 1: Dimension. It is a non-negative interval

with an associated label. The interval is deﬁned by an

ordered pair of values [d

] := {x ∈ R | d

≤ x ≤

}, with d

≤ d

. The interval lower bound d

0 for every dimension, while the upper bound d

can

vary for each dimension.

Deﬁnition 2: Range. It is a continuous semi-open

interval deﬁned on a dimension d. It is deﬁned by an

ordered pair of values [r

) := {x ∈ R | r

≤ x <

}, with r

− r

≥ 1. The component of the range

and r

are known as range lower bound and range

upper bound, respectively.

Deﬁnition 3: Region Speciﬁcation. It is deﬁned as a

set of ranges, named RS. RS must contain at most one

range r for any given dimension d ∈ D, where the di-

mension set D is derived starting from the ranges that

constitute RS. Each range r ∈ RS is deﬁned in terms

of bounds [0, d

), where d

is the corresponding di-

mension’s upper bound.

Deﬁnition 4: Region Template. It is deﬁned as an

incomplete region speciﬁcation in which one or more

dimensions have not been assigned ranges.

Deﬁnition 5: Region Realization. It is deﬁned as a

region speciﬁcation that is associated with an instance

attribute for update, a sent interaction, or a class at-

tribute or interaction class for subscription. The term

region may be used in cases where a region speciﬁca-

tion, a region realization, or both apply.

The DDM process goes through four phases: (i)

declaration, (ii) match, (iii) connect, and (iv) for-

ward, which repeatedly occur throughout the feder-

ation execution (IEEE Std. 1516-2010, 2010; Zhu

and Wang, 2022). In the declaration step, feder-

ates create regions with a speciﬁc set of dimensions,

and declare the HLAObjectClass and/or HLAInterac-

tionClass data that they intend to publish and/or sub-

scribe, in terms of update and subscription regions,

respectively. In the match phase, the overlap between

each pair of update and subscription regions is cal-

culated, and obtained pairs are reported to the RTI.

After that, in the connect phase, the RTI establishes

connections between the sending federate and receiv-

ing ones. Finally, In the forward phase, data are for-

warded through the created connections.

The region matching problem can be deﬁned as

follow. Given an update region set U = {u

,...,u

}

and a subscription region set S = {s

,...,s

}, such

that U

S ̸= {

0}, and R = U

S. An update u

and a

subscription s

region overlap if and only if all ranges

of dimensions that are contained in both regions over-

A Topic-Based Data Distribution Management for HLA

187

lap pairwise. If the regions do not have any dimen-

sions in common, they do not overlap. Algorithm 1

delineates the steps to check for overlap between two

ranges a = [a

) and b = [b

) for a given a

dimension d.

Algorithm 1: Calculation of an overlap between ranges for

a given a dimension d.

Require: d ▷ dimension

Require: a, b ▷ ranges deﬁned both on d

1: overlap ← {(a

= b

)∨(a

< b

∧ b

< a

)}

2: return overlap

Figure 1 depicts an example of a two-dimensional

region matching problem composed of two subscrip-

tion regions S = {s

} and one update region U =

} both speciﬁed on the dimensions X and Y ,

where, for example, X and Y can represent latitude

and longitude, respectively (M

oller et al., 2016). It is

easy to identify that the pair {(s

)} is the solution

to the region matching problem, since their ranges

overlap on both dimensions. The pair {(s

)} is

not part of the solution; therefore, s

will receive data

from u

while s

will not.

Dimension X

Dimension Y

Figure 1: An example of the region matching problem in a

two-dimensional space.

The correct deﬁnition of the region matching al-

gorithm is crucial for the network and processing cost

(Raczy et al., 2005). This problem can be reduced

to the rectangles intersection problem. To ﬁnd the

intersections between two-dimensional rectangles, in

(Bentley and Wood, 1980), the authors developed

an algorithm with computational complexity equal to

O(N · log(N) + K), where N is the number of two-

dimensional rectangles and K is the number of inter-

secting pairs discovered. However, the proposed al-

gorithm can only be applied to two-dimensional rect-

angles and does not offer a generic solution for n-

dimensional rectangles.

The following section reviews the existing liter-

ature contributions dealing with both the DDM ser-

vices and region matching algorithms.

3 RELATED WORK

Four relevant DDM matching algorithms have been

developed, i.e., Region-Based Matching (RBM),

Grid-Based Matching (GBM), Hybrid-Based Match-

ing (HBM), and Sort-Based Matching (SBM).

3.1 Region-Based Matching

The Region-Based Matching algorithm is also known

as Brute-Force Matching algorithm. Given a dimen-

sion d, it ﬁnds overlaps by comparing each update re-

gion with all the subscription ones. The RBM match-

ing algorithm is simple to implement and allows to

derive exact overlapping information. The compu-

tational complexity is O(n · m), where n and m are

the number of update and subscription regions, re-

spectively. However, this computational complexity

is affected by the number of dimensions; therefore,

in the generic d-dimensional case, i.e., running the

RBM algorithm for each dimension d and calculat-

ing intersections between regions, the computational

complexity is O(d · (n · m)).

The main advantage of this algorithm is its sim-

plicity, but it has scalability issues as regions and di-

mensions grow. RBM was adopted in the ﬁrst ver-

sion of the Defense Modeling and Simulation Ofﬁce

(DMSO) RTI implementation of the HLA 1.3 spec-

iﬁcations, and in the M

AK High-Performance RTI

(Wood, 2002; Pan et al., 2011).

3.2 Grid-Based Matching

The Grid-Based Matching algorithm works by par-

titioning the Multi-Dimensional Coordinate Space

(MDCS) into a grid of cells. Each region r is then

mapped, through a function f (r), to the correspond-

ing cells of the grid. In GBM, an overlap between an

update and a subscription region happens if and only

if they have at least one cell of the grid in common

(Boukerche et al., 2005).

While the GBM algorithm has a lower computa-

tional overhead than the RBM one, the overlapping

information is not exactly determined. Therefore, ir-

relevant data may be received by a subscribing feder-

ate f

even if it has no region overlaps with the up-

dating federate f

, but only because f

and f

have

at least one cell in the grid in common. Obviously,

SIMULTECH 2023 - 13th International Conference on Simulation and Modeling Methodologies, Technologies and Applications

188

this leads to unnecessary consumption of network re-

sources, and f

has to discard irrelevant data it re-

ceives. To overcome this issue, many RTI implemen-

tations with GBM-based DDM services adopt an ad-

ditional data ﬁlter on the receiving side.

The computational complexity is O(c + n · m/c),

where c is the number of cells, n and m are the num-

ber of update and subscription regions, respectively.

In the generic d-dimensional case, i.e., running the

GBM algorithm for each dimension d, the computa-

tional complexity is O(d ·(c +n ·m/c)) (Marzolla and

D’angelo, 2020).

Like RBM, this algorithm is simple to implement

but is more scalable than the ﬁrst one. The perfor-

mance of the GBM algorithm depends not only on the

grid size but also on the size of the individual cells

that compose it. Indeed, the larger the cell size, the

greater the amount of irrelevant data transferred, but

the shorter the time the GBM algorithm takes to de-

tect overlaps between regions. On the other hand, the

smaller the cell size, the less irrelevant data will be

transferred, but the GBM algorithm needs much more

computational resources to determine overlaps. For

this reason, the choice of the cell size represents a cru-

cial aspect as it impacts the performance of the DDM

services based on GBM (Ayani et al., 2000; Tan et al.,

2000a).

3.3 Hybrid-Based Matching

The Hybrid-Based Matching algorithm combines the

GBM and RBM algorithms. Speciﬁcally, GBM is

used to partition the MDCS into a grid of cells, then

map regions to grid cells; while RBM is used to per-

form exact matching between update and subscription

regions that overlap the same cells in the grid (Tan

et al., 2000b).

This approach has two advantages. On the one

hand, it has a lower computational cost than the RBM

algorithm in determining the overlap information. On

the other hand, it overcomes the issue of the GBM al-

gorithm since it allows obtaining exact overlapping

information. Nevertheless, the main issue of the

HBM algorithm is that it has the same drawbacks as

GBM, i.e., the performance depends on the size of

both the grid and cells.

3.4 Sort-Based Matching

The Sort-Based Matching algorithm improves match-

ing performance by sorting the boundaries of re-

gions before evaluating their overlap (Pan et al., 2007;

Raczy et al., 2005).

The algorithm uses a bit-matrix M ∈ R

nxm

, where

n and m are the number of update and subscription

regions, respectively. Semantically, a row n

, with i =

0,...,|n| indicates the update region, while a column

, with j = 0,...,|m| the subscription region. Each

element M

i, j

is deﬁned as follow:

(

1, if the regions i, j overlap,

0, otherwise.

(1)

Two sets of subscription regions (SubSetBe f ore and

SubSetA f ter) are deﬁned as n-bit vectors to track re-

gions. The algorithm operates as follows. Given a

dimension d, it begins by assuming that each update

region overlaps each subscriber region; as a conse-

quence, each matrix element is initialized to 1. Then,

it inserts the ranges’ bounds of all regions into an or-

dered list ord list, initializes SubSetBe f ore = {

0},

and adds all subscription regions into SubSetA f ter.

Upon the initialization is completed, ord list is

scanned from bottom to top and operates, for each

element l

, with i = |ord list|,...,0, as follows. If l

is:

• a lower bound of a subscription region R, then

SubSetA f ter \ {R};

• an upper bound of a subscription region R, then

SubSetBe f ore ∪ {R};

• a lower bound of an update region R, then all re-

gions in SubSetBe f ore do not overlap with R, then

M is updated;

• an upper bound of an update region R, all regions

in SubSetA f ter do not overlap with R, then M is

updated.

The computational complexity is O(n · m), where n

and m are the number of update and subscription

regions, respectively. In the generic d-dimensional

case, i.e., running the SBM algorithm for each di-

mension d, the computational complexity becomes

O(d · (n · m)) (Raczy et al., 2005).

4 A TOPIC-BASED

PUBLISH-SUBSCRIBE

APPROACH FOR DDM

Despite research progress, most of the existing

region-matching algorithms need to scan every region

to ﬁnd overlaps with others, resulting in a waste of

computing resources. To face these needs and short-

comings, the Topic-based publish-subscribe messag-

ing system (TBMS), of which the Topic-Based Match-

ing algorithm (TBM) is part, has been deﬁned. The

idea behind this new messaging system is to have an

A Topic-Based Data Distribution Management for HLA

189

Run-time Infrastructure (RTI)

Data Distribution Management (DDM)

Producer

Topic

Queue (1)

Topic

Queue (2)

Topic

Queue (N)

Exchanger

…

Binding

Consumer

(A)

Consumer

(B)

Update

Data

Figure 2: Architecture of the Topic-based publish-subscribe

messaging system.

“exchanger” with a queuing system that mediates the

communication between federates, minimizing mu-

tual awareness, i.e., what federates should have of

each other to be able to exchange messages, effec-

tively implementing decoupling. A producer feder-

ate sends a message to the exchanger, which in turn

forwards it, by using the TBM algorithm, to the cor-

responding queue that consumer federates use to get

the message. A key advantage of TBMS is that the ex-

changer forwards messages to queues without need-

ing to know consumer federates.

TBMS changes the way with which DDM ser-

vices manage regions and ﬁnd overlaps by intro-

ducing the concept of “topic”. A topic is a well-

structured string deﬁned using a dot-delimited format,

and it is used to ﬁlter and deliver HLAObjectClass

and HLAInteractionClass messages. The structure of

a topic is composed of three parts:

region id.{ob ject|interaction}.instance id (2)

where:

• region id represents the identiﬁer of the region;

• {ob ject|interaction} speciﬁes the supported

datatype: object, for managing HLAObjectClass

and interaction for handling HLAInteractionClass

(IEEE Std. 1516-2010, 2010);

• instance id represents the identiﬁer of the in-

stance.

Figure 2 depicts the architecture of TBMS along

with the key parts: Producer Federates, Exchanger,

Bindings, Queues, and Consumer Federates.

Producer Federates (see, Federate (A) in Figure 2)

are responsible for creation regions by using the cre-

ateRegion() method, with a speciﬁc set of dimensions.

For every dimension, the lower and upper bounds

of the range of that region are deﬁned through the

setRangeLowerBound() and setRangeUpperBound()

methods, respectively (IEEE Std. 1516-2010, 2010).

A Producer Federate sends HLAObjectClass and

HLAInteractionClass messages to the Exchanger

component. When the Exchanger receives the mes-

sage, it is responsible for routing it to different Queues

by using the message’s topic information and Bind-

ings, which connect the exchange and queues.

The wildcard “*” has been deﬁned to identify all

elements in a speciﬁc position of the topic structure.

This wildcard may be used to deﬁne a binding over a

topic following the same topic’s structure and Rule 1:

Rule 1. If the second part of the binding uses the

wildcard, then the last part must have the wildcard.

For example, < region id > . ∗ .∗, and

< region id > .ob ject.∗ are valid bindings; whereas,

< region id > . ∗ . < instance id > is invalid.

The Exchanger component forwards messages to

queues depending on wildcard matches between the

message’s topic and the queue binding’s routing pat-

tern. It is important to note that once the Producer

Federates sends a message to the Exchanger, it does

not wait for a response, i.e., it is not blocked.

Consumer Federates (see, Federates (B) and (C)

in Figure 2) deﬁnes one or more bindings and sub-

scribe to the related queues in order to receive mes-

sages from Producer Federates, and then process

them.

In order to explain the Topic-based publish-

subscribe messaging system more easily, a scenario

related to a drone patrol system is examined, as re-

ported in Figure 3. In this scenario, Federate (A) sim-

ulates three cars in the New York area, and related

data are published/updated on the RTI using the DDM

services extended with the proposed solution. Feder-

ates (B) and (C) simulate two drones used to patrol

the New York area, with the difference that Federates

(B) is interested in tracking and monitoring all cars,

whereas Federates (C) wants to track only “car1”.

When the simulation starts, Federate (A) creates

the region r

and simulates the movement of the three

cars (car

, car

, and car

) in the New York area. Fed-

erates (B) and (C) deﬁne the bindings r

.ob ject.∗ and

.ob ject.car

, respectively, and subscribe to the re-

lated queues. Every time Federate (A) updates data

related to a car and publishes it on the RTI, the fed-

erate sends a message consisting of two parts: (i) the

reference topic; and (ii) the car’s data (see Figure 3b

- step 1). Upon the Exchanger receives the message

(see Figure 3b - step 2), it uses the TBM algorithm

to forward the message to the corresponding queues

(see Figure 3b - step 3), and then to the subscriber

Federate(s) (see Figure 3b - step 4).

4.1 Topic-Based Matching Algorithm

The Topic-Based Matching algorithm allows the Ex-

changer component to deliver messages to queues

based on wildcard matches between the topic and the

binding patterns, which is speciﬁed by queues. All

SIMULTECH 2023 - 13th International Conference on Simulation and Modeling Methodologies, Technologies and Applications

190

Car1

Car2

Car3

(a)

Run-time Infrastructure (RTI)

Data Distribution Management (DDM)

Federate

(A)

Topic

Queue (1)

Topic

Queue (2)

Exchanger

Federate

(B)

Federate

(C)

r1.object.car1

r1.object.*

msg

(r1.object.car2, )

msg

(b)

Figure 3: A scenario related to a drone patrol system: (a)

depicts the positions of car1, car2, and car3 on the map;

whereas, (b) reports the conﬁguration of TBMS.

messages with a topic that matches a binding’s pat-

tern are routed to the queue, and then forwarded to

the consumer federates. Algorithm 2 presents the

steps performed by TBM to ﬁnd overlaps between the

topic and binding patterns. It uses a key − value data

Algorithm 2: The Topic-Based Matching algorithm.

Require: t, m ▷ t - topic, m - message

1: bindings = Map < binding pattern,queue >

2: for entry in bindings do

3: bind ← entry.binding pattern

4: if wildcardMatch(bind,t) then

5: q ← entry.queue

6: q.enqueue(m)

7: else

8: discard m

9: end if

10: end for

structure to manage the associations between bind-

ing patterns and queues. For each entry in bindings

(see line 2), the algorithm checks whether the wild-

card bind matches with the topic t or not (Golan et al.,

2019). If yes, the queue q is retrieved, and then the

message m is enqueued into q in order to be forwarded

to all subscribed federates (see lines 5- 6). Otherwise,

the algorithm continues with the next entry.

The “wildcardMatch” method has a computa-

tional complexity of O(m), where m is the length of

the topic (Hajiaghayi et al., 2021). Since it is called

for each entry in bindings, the computational com-

plexity of TBM algorithm is O(n · m).

The following section presents the experiments

carried out to evaluate the performance of the Topic-

Based Matching algorithm. The gathered results have

been compared with the matching algorithm adopted

in HLA 4.0 (see, Algorithm 1).

5 EXPERIMENTAL EVALUATION

This section presents the experiments carried out

to evaluate the performance of the TBM algorithm.

The experiments have been written in the Java lan-

guage and performed on a MacBook Pro, equipped

with MacOS Catalina 10.15, 16GB of RAM, and

1TB of HD. To promote comparability of the simu-

lation results with those available in the literature, the

two − dimensional case has been considered.

To perform the experiments a RabbitMQ infras-

tructure has been set up (Rostanski et al., 2014;

Ayanoglu et al., 2016). RabbitMQ is a message-

oriented middleware, also known as message-broker,

that implements the Advanced Message Queuing Pro-

tocol (AMQP). It offers a common platform for send-

ing/receiving messages, ensuring the security of com-

munications. RabbitMQ acts as an intermediary be-

tween message consumers and producers, this pecu-

liarity makes it easy to decouple the involved parts.

RabbitMQ guarantees the delivery of messages, pro-

vides non-blocking features, and can be conﬁgured to

push notiﬁcations to producers. Moreover, it provides

support to the publish/subscribe mechanism, asyn-

chronous processing, and message queues.

To conduct the experiments, three factors have

been considered: Total number of regions, Overlap

rate, and Update rate. The ﬁrst one is the most ob-

vious factor. If there are more regions, the match-

ing algorithm will take longer to determine overlaps.

The total number of regions, used in the experiments,

varies from 1000 to 10000. Each region has a size

of 20x20. All regions are evenly distributed in a

5000x5000 routing space.

The Overlap rate represents, as deﬁned by Equa-

tion 3, the number of subscribed regions over the total

number of regions. The higher this rate, the longer it

takes for the algorithm to check whether or not two

regions intersect. In the experiments, three different

overlap rates were used: 0.25 (low), 0.50 (medium),

and 0.75 (high). Finally, the Update rate represents

A Topic-Based Data Distribution Management for HLA

191

the generation rate of update events on regions de-

ﬁned by using a probabilistic model. It is used by the

producer federate to rate the generation of messages

to subscriber federates. In the experiments, the Pois-

son distribution has been chosen with mean number

of events per time interval λ = 0.8 in order to avoid

an excessive generation of events.

Overlap rate =

|subscribed regions|

|total regions|

(3)

Figure 4 shows the time performance comparison be-

tween T BM and RBM matching algorithms with a low

overlap rate, i.e., 0.25. In this situation, since there

is a low number of intersections between subscription

and update regions, the computational cost for ﬁnding

overlaps does not vary much between the two consid-

ered algorithms as long as the total number of regions

remains relatively low, i.e., less than 6000 regions.

While, with a large number of regions, more signif-

icant than 6000, the T BM algorithm achieves much

better performance than RBM.

0 2000 4000 6000 8000 10000

Number of regions

Time (ms)

TBM

RBM

Figure 4: Performance comparison between the TBM and

RBM matching algorithms, with low overlap rate 0.25.

Figure 5 depicts the time performance compari-

son between the TBM and RBM matching algorithms

with an overlap rate of 0.50. In this situation, the com-

putational cost of both matching algorithms clearly

increases with respect to the previous case, since there

are more regions to evaluate. However, similar to

the previous scenario, the performance trend remains,

and the performance of the RBM algorithm is very

poor, while the proposed T BM algorithm has a better

processing time, especially when the total number of

regions is greater than 6000.

Concerning the scenario with a high overlap rate,

i.e., 0.75, there is a high degree of intersections be-

tween regions. The computational cost of both match-

ing algorithms still increases compared to the previ-

ous cases, as both directly depend on the total num-

ber of regions. Similarly to the previous scenarios,

0 2000 4000 6000 8000 10000

100

125

150

Number of regions

Time (ms)

TBM

RBM

Figure 5: Performance comparison between the TBM and

RBM matching algorithms, with medium overlap rate 0.50.

the computational cost shows the same trend, and the

RBM algorithm keeps achieving a lower performance

than the proposed T BM one. Figure 6 shows the time

performance comparison between the T BM and RBM

matching algorithms with a high overlap rate.

0 2000 4000 6000 8000 10000

500

1000

1500

2000

Number of regions

Time (ms)

TBM

RBM

Figure 6: Performance comparison between the TBM and

RBM matching algorithms, with high overlap rate 0.75.

6 CONCLUSIONS

In this paper, the novel Topic-based publish-subscribe

messaging system (TBMS), of which the Topic-Based

Matching algorithm (TBM) is part, has been deﬁned

to improve the performance, reliability, and scalabil-

ity of DDM services in HLA. To evaluate the perfor-

mance of T BM, a set of experiments has been carried

out by considering different overlap rates. The exper-

iment results were compared with the ones obtained

with the standard Region-Based Matching (RBM) al-

gorithm. Results highlight the fact that the proposed

T BM algorithm achieves better performance than the

RBM one in the all considered overlap rates. Further

investigation will focus on the reliability and scalabil-

ity of the proposal.

SIMULTECH 2023 - 13th International Conference on Simulation and Modeling Methodologies, Technologies and Applications

192

REFERENCES

Ayani, R., Moradi, F., and Tan, G. (2000). Optimizing

cell-size in grid-based ddm. In Proceedings Four-

teenth Workshop on Parallel and Distributed Simula-

tion, pages 93–100. IEEE.

Ayanoglu, E., Aytas, Y., and Nahum, D. (2016). Mastering

rabbitmq. Packt Publishing Ltd.

Bentley, J. L. and Wood, D. (1980). An optimal worst

case algorithm for reporting intersections of rectan-

gles. IEEE Transactions on Computers, 29(07):571–

577.

Boukerche, A., McGraw, N. J., Dzermajko, C., and Lu, K.

(2005). Grid-ﬁltered region-based data distribution

management in large-scale distributed simulation sys-

tems. In 38th Annual Simulation Symposium, pages

259–266. IEEE.

Bouskela, D., Falcone, A., Garro, A., Jardin, A., Otter,

M., Thuy, N., and Tundis, A. (2021). Formal Re-

quirements Modeling for Cyber-Physical Systems En-

gineering: an integrated solution based on FORM-L

and Modelica. Requirements Engineering, 27(1):1–

30.

Derler, P., Lee, E. A., and Sangiovanni Vincentelli, A.

(2012). Modeling cyber–physical systems. Proceed-

ings of the IEEE, 100(1):13–28.

Falcone, A., Garro, A., Mukhametzhanov, M. S., and

Sergeyev, Y. D. (2020). Representation of Grossone-

based Arithmetic in Simulink for Scientiﬁc Comput-

ing. Soft Computing, 24(23):17525–17539.

Falcone, A., Garro, A., Mukhametzhanov, M. S., and

Sergeyev, Y. D. (2022). Simulation of Hybrid Systems

Under Zeno Behavior Using Numerical Inﬁnitesimals.

Communications in Nonlinear Science and Numerical

Simulation, 111:106443.

Falcone, A., Garro, A., and Tundis, A. (2014). Mod-

eling and simulation for the performance evaluation

of the on-board communication system of a metro

train. In the 13th International Conference on Mod-

eling and Applied Simulation, MAS 2014, Held at the

International Multidisciplinary Modeling and Simula-

tion Multiconference, I3M 2014, Bordeaux, France,

September 10-12, 2014, pages 20–29. Dime Univer-

sity of Genoa.

Fujimoto, R. M. (2000). Parallel and distributed simulation

systems, volume 300. Citeseer.

Golan, S., Kopelowitz, T., and Porat, E. (2019). Stream-

ing pattern matching with d wildcards. Algorithmica,

81(5):1988–2015.

Hajiaghayi, M., Saleh, H., Seddighin, S., and Sun, X.

(2021). String matching with wildcards in the mas-

sively parallel computation model. In Proceedings

of the 33rd ACM Symposium on Parallelism in Algo-

rithms and Architectures, pages 275–284.

IEEE Std. 1516-2010 (2010). IEEE Standard for Model-

ing and Simulation (M&S) High Level Architecture

(HLA): 1516-2010 (Framework and Rules); 1516.1-

2010 (Federate Interface Speciﬁcation); 1516.2-2010

(Object Model Template (OMT) Speciﬁcation).

Lee, E. A. (2008). Cyber physical systems: Design chal-

lenges. In 2008 11th IEEE International Symposium

on Object and Component-Oriented Real-Time Dis-

tributed Computing (ISORC), pages 363–369.

Marzolla, M. and D’angelo, G. (2020). Parallel data distri-

bution management on shared-memory multiproces-

sors. ACM Transactions on Modeling and Computer

Simulation (TOMACS), 30(1):1–25.

oller, B., Antelius, F., Johansson, M., and Karlsson, M.

(2016). Building scalable distributed simulations: De-

sign patterns for hla ddm. In Proc. of Fall Simulation

Interoperability Workshop, 2016-SIW, volume 3.

Pan, K., Turner, S. J., Cai, W., and Li, Z. (2007). An ef-

ﬁcient sort-based ddm matching algorithm for hla ap-

plications with a large spatial environment. In 21st

International Workshop on Principles of Advanced

and Distributed Simulation (PADS’07), pages 70–82.

IEEE.

Pan, K., Turner, S. J., Cai, W., and Li, Z. (2011). A dynamic

sort-based ddm matching algorithm for hla applica-

tions. ACM Transactions on Modeling and Computer

Simulation (TOMACS), 21(3):1–17.

Raczy, C., Tan, G., and Yu, J. (2005). A sort-based

ddm matching algorithm for hla. ACM Transactions

on Modeling and Computer Simulation (TOMACS),

15(1):14–38.

Rostanski, M., Grochla, K., and Seman, A. (2014). Evalua-

tion of highly available and fault-tolerant middleware

clustered architectures using rabbitmq. In 2014 feder-

ated conference on computer science and information

systems, pages 879–884. IEEE.

Tan, G., Ayani, R., Zhang, Y., and Moradi, F. (2000a). Grid-

based data management in distributed simulation. In

Proceedings 33rd Annual Simulation Symposium (SS

2000), pages 7–13. IEEE.

Tan, G., Zhang, Y., and Ayani, R. (2000b). A hybrid

approach to data distribution management. In Pro-

ceedings Fourth IEEE International Workshop on Dis-

tributed Simulation and Real-Time Applications (DS-

RT 2000), pages 55–61. IEEE.

Wood, D. D. (2002). Implementation of ddm in the mak

high performance rti. In Proceedings of the simulation

interoperability workshop. Citeseer.

Zhu, G. and Wang, H. (2022). Empirical study of

large-scale hla simulation of parallel region-matching

knowledge recognition algorithm based on region

matching. Computational Intelligence and Neuro-

science, 2022.

A Topic-Based Data Distribution Management for HLA

193