Extending UML/MARTE-SAM for Integrating Adaptation

Mechanisms in Scheduling View

Mohamed

Naija and Samir Ben Ahmed

Laboratory of Computer for Industrial Systems, INSAT, Tunis, Tunisia

Keywords: Adaptability, Scheduling Analysis, Real-time & Embedded Systems, MDE, MARTE.

Abstract: The profile for Modeling and Analysis of Real-Time and Embedded systems (MARTE) defines a framework

for annotating non-functional properties of embedded systems. In particular, the SAM (Schedulability

Analysis Model) sub-profile offers stereotypes for annotating UML models with the needed information

which will be extracted to fulfil a scheduling phase. However, SAM does not allow designers to specify data

to be used in the context of adaptive systems development. It is in this context that we propose an extension

for the MARTE profile, and especially the sub-profile Schedulability Analysis Modeling, to include

adaptation mechanisms in scheduling view. We illustrate the advantages and effectiveness of our proposal by

modeling a FESTO case study as an Adaptive Real-Time and Embedded system.

1 INTRODUCTION

The modeling of Real-time & Embedded Systems

(RTES) may be stated as a crucial problem in the

software engineering domain. RTES are subject to a

multitude of constraints (e.g., battery, temperature

…) and real-time requirements. Thus, designers are

encountering the challenge of resource limitations,

time, highly variable environment, etc. The addition

of adaptivity to such systems further hardens and

delays their modelling and scheduling analysis

especially with the current lack of design models and

tools for adaptive RTES. Lightening the task of

adaptive systems designers and reducing the

development cost and time to market represent a

major challenge in the field, which requires the use of

high-level approaches such as MDE (Schmidt, 2006)

and MARTE (OMG, 2008).

MDE is a way to beat the growing complexity of

real time systems and verifying their correctness. In

particular, Unified Modeling Language (UML)

profiles promote an adequate solution to support the

whole lifecycle co-design of complex systems. In

RTES domain, its adoption is seen promising for

several purposes: requirements specification,

behavioral and architectural modeling with their real

time constraints and performance issues. In this

context, the profile for Modeling and Analysis of

Real-Time and Embedded systems (MARTE) fosters

the building of models that support the specification

of scheduling analysis problem. This profile has the

capacity to model tasks, dependencies between them

and events under shape a Workload Behavior of

system. Subsequently, it promotes the validation of

the system temporal accuracy. Unfortunately,

MARTE does not define a clear semantics for

modeling and analysis of the adaptation in RTES.

Thus, we propose in this paper the main changes

to be made on MARTE profile for supporting

adaptation mechanisms. These amendments affect

mainly the stereotypes of MARTE/SAM (Scheduling

Analysis Modeling) since it is the sub-profile

intended to model the schedulability analysis.

Our contribution is to improve the meta-models of

the existing annotations. We try to modify in the

structure of existing annotations, with no need to add

new ones, by referring to their cardinality which

makes easy the adoption of the proposed extension in

revision of MARTE. This work is the result of

previous investigations and published work. Starting

from (Naija et al., 2015) a MARTE-based approach

was proposed to concurrency model construction at

early design stages. In (Naija and Ben Ahmed, 2015)

a reconfiguration solution for RTES to meet

performance constraints is developed. Notably, we

perceive a MARTE semantics limitation in modeling

level. Accordingly, we have identified the needed of

the new version of the MARTE profile supporting

adaptation mechanisms.

This work is to be integrated in a model-based

Naija M. and Ahmed S.

Extending UML/MARTE-SAM for Integrating Adaptation Mechanisms in Scheduling View.

DOI: 10.5220/0005822300840090

In Proceedings of the 11th International Conference on Evaluation of Novel Software Approaches to Software Engineering (ENASE 2016), pages 84-90

ISBN: 978-989-758-189-2

approach to guide RTES designers for building and

analysing adaptive RTES models. It facilitates

complex systems modeling, reduces the development

time and cost and improves software process quality.

The above benefits have been illustrated through the

application of our extensions to different examples of

adaptive RTES.

The present paper is organized as follows. Section

2, emphasizes the various related works, section 3

introduces the concept of the adaptation. While

section 4 gives a brief definition of the MDE

paradigm, its MARTE profile and the SAM sub-

profile, section 5 specifies our proposal. To better

explain our contribution which is highlighted in

section 6, we rely on a case study. Finally, section 7

concludes the paper and sketches some future work.

2 RELATED WORK

The design of adaptive RTES presents many

challenges due to the complexity of the problem it

handles (Said et al., 2014). In the present paper, we

limit our study to research works particularly tackling

adaptive RTES using the MARTE profile.

Many researchers have benefited from the

MARTE profile for the design and verification of

adaptive RTES from high-level models. In (Cherif et

al., 2010) authors have benefited from MARTE to

model reconfigurable architectures such as FPGAs

based Systems-on-Chip (SOC). They extended the

MARTE profile with some semantics and Xilinx

specific concepts, which limits their applicability for

diverse systems, to support Dynamic and Partial

Reconfiguration (DPR) of FPGA. Unlike this

contribution, we aim to propose a new extension to

support adaptation which is independent from any

specific platform.

In (Boukhanoufa et al., 2010) the authors give a

classification of 13 publications that have dealt with

the subject of adaptation in the design approach.

Following this classification, the authors illustrate

using an avionic example the need for the validation

of adaptation rules at design-time according to the

real-time features of the system. In this context of

verification approaches, they have proposed in

(Boukhanoufa et al., 2011) an MDE approach for

modeling and offline validation of application timing

constraints. In fact, this article uses state machine to

represent the application configurations and

transitions between them to represent adaptation

rules. This work is based on the generation of all

possible configurations of a system before running, in

order to validate timing constraints. The number of

configurations varies from one system to another and

it can be very large, this combinatorial explosion

makes the timing analysis inapplicable.

In (Said et al., 2014), five patterns have been

proposed to model and evaluate adaptation. These

design patterns are presented in a static form through

class diagrams and stereotyped MARTE profile. The

adaptive behavior of the system is then designed

according to the proposed design pattern without

supporting schedulability analysis.

Two major scheduling approaches are available in

the literature (Magdich et al., 2014): the partitioned

and the global approaches. Originally, MARTE

supports only the modeling of the systems to be

scheduled according to the partitioned approach. In

(Magdich et al., 2012) the authors have proposed

various updates for MARTE meta-models of

specialization and generalization stereotype in order

to support global scheduling approaches, allowing

task migrations. Those changes allow a schedulable

resource to be executed on different computing

resources in the same period (Rubini et al., 2014).

Unfortunately adaptation is considered only as a

dynamic change in the task allocation, without taking

into account operating mode adaptation which is an

essential axe of adaptation.

3 THE ADAPTABILITY

DEFINITION

There are several definitions of adaptation in the

literature. In (Bihari, 1991), a software adaptation is

defined as any software modification that changes the

reliability or timeliness of the software without

affecting other aspects of its functionality. Software

adaptation encompasses many common software-

tuning techniques. These include:

• resource reallocation, such as moving a software

component from one processor to another,

• adjustments to processor schedules,

• modification of replication factors for N-modular-

redundant software components, and

• modification of retry limits or time-out periods for

delivery of a service by a software component.

In (Subramanian and Chung, 2000) and (Lehman and

Ramil, 2000), adaptation means change in the system

to accommodate change in its environment. More

specifically, the adaptation of a software system (S)

is caused by a change (Che) from an old environment

(E) to a new environment (E’), and results in a new

system (S’) that ideally meets the needs of its new

Extending UML/MARTE-SAM for Integrating Adaptation Mechanisms in Scheduling View

environment (E’). Formally, adaptation can be

viewed as a function:

Adaptation: E × E’ × S→ S’, where meet (S’,

need (E’)).

In (Oreizy et al., 1999) adaptive system is defined

as a system that is able to change its structure or

behavior at run-time in response to the execution

context variations and according to adaptation engine

decisions.

In the present paper, adaptation is defined as any

modification in the structure, behavior or architecture

of the system to accommodate external or internal

change of their operating environment or context and

according to predefined adaptation plan and rules.

We distinguish two types of reconfigurations; static

reconfigurations (Angelov and Marian., 2005) and

dynamic reconfigurations (Khalgui and Hanisch,

2011). Since static adaptation requires stopping the

system and restarting it with a new configuration. The

dynamic adaptation is based on a set of predefined

adaptation behavior, statically designed and verified at

design time. At runtime, the system can select an

alternative from the available ones in accordance with

adaptation rules and context.

4 MDE AND RTES

DEVELOPMENT

The Model Driven Engineering (MDE) is a software

development methodology aiming to increase the

level of development and overcome the growing

complexity challenge. It covers the entire systems

lifecycle, simplifies the design process by using the

concept of models and offers independency between

different steps of development flow.

In the context of the schedulability analysis, MDE

is mainly used in the modeling step and the

transformation of scheduling analysis models to the

models of the chosen scheduling analysis tool

(Magdich et al., 2012). MDE uses the UML profile

and especially MARTE.

4.1 MARTE Capabilities for RTES

Modeling

MARTE is an extension of UML profile providing

support for specification, modeling and verification

step of real time and embedded systems. MARTE

supports the modeling of software and hardware

features at a high-level of abstraction. In addition, it

offers a rich set of annotations for modeling

schedulability analysis. This profile encompasses a lot

of sub-profiles such as: SRM (Software Resource

Modeling), HRM (Hardware Re-source Modeling),

GQAM (Generic Quantitative Analysis Modeling),

SAM (Schedulability Analysis Modeling), PAM

(Performance analysis Modeling), etc. In this paper, we

will focus especially on the SAM sub-profile since it is

the package affected by our proposal.

4.2 SAM

In order to establish an early validation of the

system’s temporal behavior a well-formed analyzable

model, called SAM, is defined. This profile has the

capacity to model tasks, dependencies between them

and events. It has the capacity to predict if all tasks

meet their time constraints by defining the workload

behavior. This is a chain of operation activations

representing executions scenarios for the application.

5 SAM EXTENSION FOR A

BETTER SCHEDULING

ANALYSIS MODELING STEP

The scheduling analysis modeling is performed

through the MARTE/SAM profile. The idea of

performing scheduling analysis based on MARTE

models assumes that all the information that is needed

for the analysis is already part of the MARTE model

(Naija et al., 2015). In fact, SAM meta-model supports

the modeling of different systems as it models all the

temporal features needed in the scheduling step except

those used to model adaptation constraints.

Consequently, we seek to improve SAM meta-model

in order to support modeling and early analysis of

adaptation process. The amendments to be done on the

SAM sub-profile affects also the GQAM sub-profile

since some classes of the sub-profile SAM inherit from

GQAM sub-profile.

5.1 The Workload Behavior

The end-to-end computation represents the

processing load of the system. It represents the

different steps (operations) executed in the system

and triggered by one or more external stimulus. Each

step can be linked to a successor step with a control

flow. «SaStep» is a stereotype annotating an

action/operation and contains both timing description

(computational budget, activation event, etc.) and

timing constraints (operation deadlines). A set of

steps specify the so-called Schedulable Resources.

These are units of execution taken into account by the

ENASE 2016 - 11th International Conference on Evaluation of Novel Software Approaches to Software Engineering

scheduler of the system, called tasks in scheduling

literature (Radermacher et al., 2010).

5.2 Amendments to be Done for the

Workload Event

The workload behavior of the system (OMG, 2008) is

characterized by their workload events and behavior

scenarios. Workload events annotating UML

AcceptEventActions introduce the semantic of event

sequence arrivals for the execution of each

callBehaviorAction. Originally event triggers only

one behavior scenario. When adaptation is required,

an event triggers all the behavior scenarios of the

system workload. For example, let us imagine that for

an adaptive event that denoted a low level of battery,

this can affect more than one behavior scenario. Thus,

the designer is not able to specify this property.

Hence, we propose to modify in the association

linking the two classes WorkloadEvent and

BehaviorScenario of the package GQAM Resources

(Fig 1 and Fig 2). The multiplicity [1..*] denotes that

an event can affect one or more behaviour scenarios.

Indeed, GQAM is a generalization of the package

SAM. So, this change will be inherited by SAM.

Figure 1: The old meta-model of the GQAM package.

Figure 2: The new meta-model of the GQAM package.

5.3 Amendments to be Done for the

Step

The class Step may represent a small segment of code

execution (OMG, 2008). It contains a lot of attributes

specifying the temporal features of software

resources. In UML MARTE model step can be part

of only one behavior. Otherwise, in the case of

adaptation process a new Step can appear in multiple

behavior scenarios to manage adaptation. Thus,

MARTE/SAM doesn’t allow this specification.

Thereby, we propose to change in the association

linking the two classes BehaviorScenario and Step

(Fig 4).

Figure 3: The old meta-model of the SAM package.

Figure 4: The new meta-model of the SAM package.

Once the workload behavior is performed, it is

necessary to identify the so-called schedulable

resources (called tasks in scheduling literature).

Schedulable resources are defined by mapping the

execution of the end-to-end computations to them, in

order to generate the task model. Different types of

mapping exist in the literature (Masse et al., 2003)

(Mraidha et al., 2011) (Naija et al., 2015). As

explained previously, a software resource can be

mapped into more than one thread. Accordingly, the

cardinality of the association between Step and

SchedulableResource must be [0..*] (Fig 4).

5.4 Amendments to Model Tasks

Migration

In literature, three scheduling approaches are presented:

the partitioned, the semi-partitioned and the global

approaches (Muhammad et al., 2011). Regarding the

partitioned approach, it affects each task to be executed

on one processor. Accordingly, tasks are not allowed to

Extending UML/MARTE-SAM for Integrating Adaptation Mechanisms in Scheduling View

migrate between processors (Magdich et al., 2014). CPU

utilization is therefore not optimal. As for the global

approach and semi-partitioned, they enable a tasks

migration such that schedulable resource may be

allocated, not simultaneously, on different computing

resources. The task migration is considered as an

adaptive technique that allows improving application

performance and achieves optimality. Currently,

MARTE/SAM supports only the partitioned approach.

Thus, task that can be across multiple processors for

different periods of time is not permitted in SAM.

Subsequently, the multiplicity of the attribute

corresponding to the execution resource

(ExecutionHost) on which a task (schedulableResource)

is allocated must be [0..*] instead of [0..1] (Fig 5 and Fig

6). This extension is adopted from the research work

(Magdich, 2013).

Figure 5: Meta-model of the GQAM package (Magdich,

2013).

Figure 6: Meta-model of the GQAM package with

amendments (Magdich, 2013).

While migrating from one processor to another, the

execution time of a task is not the same, then the

attribute «deadline» of the stereotype SaStep should

have a multiplicity of [0..*]. In the same vein, a task

can be interrupted several times during one period.

Consequently the attribute «preemptT», which refers

to the length of time that the step is preempted, must

have a multiplicity [0..*] instead of [0..1]. Similarly

for the attribute «readyT» which indicate length of

time since the beginning of a period. Hence, this

attribute must have a multiplicity of [0..*]. The set of

values for the attributes «deadline», «preemptT» and

«readyT» must be ordered (Fig 7 and Fig 8). This

extension is adopted from the research work (Magdich,

2012).

Figure 7: Meta-model of the GQAM package (Magdich,

2012).

Figure 8: Meta-model of the GQAM package with

amendments (Magdich, 2012).

6 CASE STUDY

To better explain our proposal, we use a FESTO

(Khalgui and Hanisch, 2011) production system as an

intact running application in this paper. It is a well-

documented laboratory system used by many

universities for research and education purposes.

The working process of FESTO is composed of

three units: the distribution unit, the test unit, and the

processing unit. The distribution unit consists of two

steps: a pneumatic feeder and a converter. It forwards

cylindrical workpieces from a stack to the testing unit.

The test unit consists of three steps: the detector, the

tester, and the evacuator. It performs the checking of

ENASE 2016 - 11th International Conference on Evaluation of Novel Software Approaches to Software Engineering

workpieces for their height, material type, and color.

Workpieces that pass the test unit successfully are

forwarded to the rotating disk of the processing unit,

where the drilling of workpieces is done. The result

of the drilling operation is next checked by a checker

and finally the finished product is removed from the

system by an evacuator.

Note that in this work two drilling machines

Drill1 and Drill2 are used to drill workpieces. Drill1

is used in case of medium production. When high

production is required, Drill2 is recommended.

According to user requirements, the system FESTO

is able to reconfigure automatically at run-time in

response to any changed working environment

caused by errors or new requirements to improve

system performance without a halt. The workload

behaviour in Fig 9 represents the processing load of

the system, founded on our proposal.

After identifying the behavior model of the

system, it is necessary to specify the so-called

schedulable Resources. For sake of simplicity, we use

in this paper the scenario-based mapping (Masse et

al., 2003) which is also one of the most used. The idea

is to regroup all the operations executed at the same

rate and belonging to the same linear end-to-end

computation to the same task. In our FESTO system,

we obtain three different threads namely task1

(pieceEjection, Convert, Test and Evacuate), task2

(pieceEjection, Convert, Test, Elevate, Rotate, Drill1,

Checker and Evacuate) and task3 (pieceEjection,

Convert, Test, Elevate, Rotate, Drill2, Checker and

Evacuate).

Following this case study, founded on our

proposal. We illustrate that an event (e.g.,

epieceEjection) triggers all tasks. The steps Convert

and Test are part of three tasks. Similar for steps

Elevate, Rotate, Checker and Evacuate, those

participate for the execution of both schedulable

resources Task2 and Task3. Compared to original

version of MARTE, specifying these properties is not

permitted. Note that, tasks can have dynamic

properties (readyT, preemptT and deadlines) due to

the concept of task migration, but we use the same

corresponding values to facilitate our example.

Anyway, we can add the different values and they

will be ordered to perform scheduling analysis. After

scheduling all tasks, we can specify in our SAM view

the used allocations through the attributes Host:

GaExecHost and the corresponding attribute ExecT:

NFP_Duration.

7 CONCLUSIONS

In this paper, we have proposed an extension to

MARTE profile since it does not allow modeling and

analysis of adaptive systems, which is judged as a

hard engineering task. This improvement affects

mainly the sub-profile MARTE/SAM and especially

Figure 9: The workload behavior of FESTO system.

Extending UML/MARTE-SAM for Integrating Adaptation Mechanisms in Scheduling View

GQAM/SAM. This contribution makes MARTE able

to stand adaptability at early design stages of RTES.

The benefit of our approach is the ability to model

adaptive properties which will be extracted, to serve

during the scheduling step. Our proposal has already

been performed on the papyrus tool, which is an

editor of MARTE-based modeling, and validated

through a case study.

As future work, we will investigate in analysing

and validating of dynamic adaptability, in accordance

with adaptation rules and context, at early design

stages. This verification step further reduces the

development risks of Adaptive RTES.

REFERENCES

Angelov C, and Marian, N., 2005. Design models for

reusable and reconﬁgurable state machines. In Proc of

Embedded Ubiquitous Comput.

Bihari T-E et Schwan K., 1991. Dynamic adaptation of real-

time software. ACMTrans. Comput. Syst., vol. 9, pp.

143-174.

Boukhanoufa M-L, Radermacher A and Terrier F., 2010.

Towards a model-driven engineering approach for

developing adaptive real-time embedded systems. In

New Technologies of Distributed Systems, pp 261-266.

Boukhanoufa M-L, Radermacher A and Terrier F., 2011.

Offline validation of real-time application constraints

considering adaptation rules. In international

Conference on Trust, Security and Privacy in

Computing and Communications, pp 974–980.

Cherif S, Quadri I.R, Meftali S, and Dekeyser J-L., 2010.

Modeling reconfigurable Systems-on-Chips with UML

MARTE profile: an exploratory analysis. Proceedings

of the 13th Euromicro Conference on Digital System

Design, pp 706-713.

Khalgui M and Hanisch H. M., 2011. Automatic NCES-

based specification and SESA-based verification of

feasible control components in benchmark production

systems. In international Journal of Modelling,

Identification and Control, vol. 12, no. 3, pp. 223–243.

Magdich A,. Kacem Y. H, Mahfoudhi A., 2013. Extending

UML/MARTE-GRM for Integrating Tasks Migrations

in Class Diagrams. In international Conference on

Software Engineering Research, Management and

Applications, pp73-84.

Magdich A,. Kacem Y. H, Mahfoudhi A, and Abid M.,

2012. A MARTE extension for global scheduling

analysis of multiprocessor systems. In international

Symposium on Software Reliability Engineering, pp

371–379.

Magdich A., Kacem Y. H., and Kerboeuf M., 2014. A

uml/marte-based design pattern for semi-partitioned

scheduling analysis. In international Workshops on

Enabling Technologies: Infrastructures for

Collaborative Enterprises, pp. 300–305.

Masse, J., Kim, S., and Hong, S. 2003. Tool Set

Implementation for Scenario-based Multithreading of

UML-RT Models and Experimental Validation. In

RTAS’ 03, Proceedings of the 9

IEEE Real-Time and

Embedded Technology and Applications Symposium.

IEEE Computer Society.

Mraidha, C., Tucci-Piergiovanni, S. and Gerard, S., 2011.

Optimum: a marte-based methodology for

schedulability analysis at early design stages. In ACM

SIGSOFT Software Engineering Notes, 36(1):1–8.

Muhammad K.B., Cécile B., and Michel A., 2011. Two

level hierarchical scheduling algorithm for real-time

multiprocessor systems. Journal of software, vol. 6, no.

11, pp. 2308–2320.

Naija M, Ahmed B.S. and, Bruel J-M., 2015. New

schedulability analysis for real-time systems based on

MDE and petri nets model at early design stages. In

ICSOFT’15. 10th International Conference on Soft-

ware Engineering and Applications, pp 330-338. 2015.

Naija M, Ahmed B.S., 2015. Aid to design and

reconfiguration of the MPSOC architectures. In

CIE’15. IEEE Computer Society 45 th International

Conference on Computers & Industrial Engineering.

Lehman M and Ramil J., 2000. Towards a Theory of

Software Evolution _ And its practical impact (working

paper). Invited Talk, In Proceedings International

Symposium on Principles of Softw. Evolution, pp 2-11.

OMG Object Management Group. 2008. A UML Proﬁle for

MARTE: Modeling and Analysis of Real-Time

Embedded systems, Beta2, Object Management Group.

Oreizy P, Gorlick M. M., Taylor R. N. et al., 1999. An

architecture based approach to self-adaptive software.

In IEEE Intelligent Systems and Their Applications,

vol. 14, no. 3, pp. 54–62.

Radermacher, A., Mraidha, C., Tucci-Piergiovanni, S. and

Gérard, S., 2010. Generation of schedulable real-time

component implementations. In ETFA’10, 15th IEEE

International Conference on Emerging Technologies

and Factory Automation. IEEE Computer Society.

Rubini S, Fotsing C, Dissaux P, Singhoff F, and Tran H. N.,

2014. Scheduling analysis from architectural models of

embedded multi-processor systems. ACM SIGBED

Review, vol. 11, no. 1.

Said M, Kacem Y.M, M. Kerboeuf M, Amor N.B. and Abid

M., 2014. Design patterns for self-adaptive RTE

systems specification. In International Journal of

Reconfigurable Computing, vol., no. 8.

Schmidt D-C. Model-driven engineering. 2006. IEEE

Computer, 39(2).

Subramanian N, Chung L., 2000. Architecture – Driven

Embedded Systems Adaptation for Supporting

Vocabulary Evolution. Proceedings of International

Symposium Principles of Software Evolution

International, vol. 0.

ENASE 2016 - 11th International Conference on Evaluation of Novel Software Approaches to Software Engineering