TRANSFORMATION OF UML DESIGN MODEL INTO

PERFORMANCE MODEL

A Model-Driven Framework

Ramrao Wagh, Umesh Bellur, Bernard Menezes

Indian Institute of Technology, Bombay, Powai, Mumbai, India

Keywords: Software Performance, Model Transformation, UML, MDA.

Abstract: Software Performance Engineering is receiving increasing attention in today’s software dominated world.

Compared to research work in performance evaluation in hardware and networks, this field is still in its

nascent stage. Many methods have been proposed but majority of them are unable to adapt in the software

development life-cycle dominated by professionals without substantial performance engineering

background. We propose UPE - a Model Driven Software Performance Engineering Framework to facilitate

performance engineering within software development life cycle, based on OMG’s MDA initiative.

1 INTRODUCTION

For a large and complex software systems such as

Enterprise systems designed using distributed

technology, it is of paramount importance that the

stakeholders are satisfied that the proposed system

will meet necessary functional as well as non

functional requirements. As development of such

systems is costly it is necessary to get a realistic

estimate of the level of quality requirements likely to

be achieved, before the resources are committed for

the project. The best possible time to make reliable

estimates that are dependable, especially for

performance, is at architectural design stage since

performance in terms of response time of software,

throughput and device utilization is dependent on the

behavioral as well as structural aspects of the

software design model.

Since UML(Infra, 2003)(Super, 2003) is being

increasingly used to express design, our overall

research work is aimed at devising performance

analysis methods applicable to software design

models expressed using UML artifacts.

This paper is structured as follows. In section 2

we look at the related work and the limitations of

this work; Section 3 then presents our solution

framework UPE that is based on MDA, section 5

will present the conclusion.

2 RELATED WORK

Over the last decade there has been a steady rise in

the number of performance prediction methods for

software systems at design stage (WOSP,98-05).

Majority of these methods follow an approach where

a performance model is constructed for the target

software and later evaluated. All these methods are

broadly based on at least one of the following

performance models: Queuing Network Model, Petri

Nets, Process Algebra, Markov Chains, Simulation

(Balsamo,2004). One of the major drawbacks of

majority of these methods is that one has to

manually construct the performance model by

studying the software system. As average software

developer is not conversant with performance

modeling, a specialist may be needed to carry out

the performance modeling.

Although number of UML based methods have

been proposed for the performance estimation from

software design (Balsamo, 2004), they still suffer

from the following major drawbacks in general:

Restricted to use of particular set of UML

diagrams

There is no proper way of specifying

performance parameters

No provision to represent the performance results

back into the model

Transformation methods are not based on

standards

They are tailored towards use of specific, often

proprietary performance analysis tools

576

Wagh R., Bellur U. and Menezes B. (2006).

TRANSFORMATION OF UML DESIGN MODEL INTO PERFORMANCE MODEL - A Model-Driven Framework.

In Proceedings of the Eighth International Conference on Enterprise Information Systems - ISAS, pages 576-579

DOI: 10.5220/0002469605760579

 SciTePress

Make use of UML designs that are stored in

lengthy and verbose XML form.

We will briefly mention three recent important

methods.

The Core Scenario Model (CSM) (Petriu, 2004)

extracts the relevant performance information

scattered implicitly and explicitly around various

UML diagrams into a single CSM which will then

facilitate the conversion into any desired

performance model that can be executed by

performance tools. Two transformations are required

i.e. from UML model to CSM (U2C) and from CSM

to Performance model (C2P). Authors propose to

use QVT(QVT, 2003) for these transformations but

due to non-availability of tools are currently using

XSLT based transformations.

(Skene, 2003) apply the Model driven

performance analysis to a distributed middleware

based enterprise application developed using EJBs.

They use MDA supported lightweight extension

mechanism using different UML profiles and

merging them to represent the design model.

Specifically UML profile for SPT and EJB are

merged to define a joint profile. A simple profile to

represent queuing network model is specified as

Analysis profile. Lack of tool support has been cited

as the reason for not automating the transformations

defined.

(D’Ambrogio, 2005) proposes a comprehensive

framework for achieving interoperability amongst

various tools based on MDA approach. Considering

the lack and shortcomings of existing tools and

specifications, they prescribe a three layer approach

comprising of a technology independent

metamodeling layer (MML), technology specific

model implementation layer (MIL) and tool layer

(TL) to exploit use of available technologies and

tools built on MDA and XML. Due to lack of proper

tool support for implementing QVT transformations,

XML based technologies such as XSLT/XQuery is

used for transformation.

Many of the drawbacks mentioned earlier are

still retained in these three representative methods.

More specifically, Out of these three methods, the

one proposed by D’Ambrogio looks most promising

as a complete but complex framework is defined by

them to achieve transformation by using proposed

standards as well as combining it with existing

technologies. However, their target performance

model is not generic. On the other hand the idea of

representing an intermediate performance model in

CSM is also promising as it combines all the

performance related information in one place before

a complex transformation is applied. Approach by

Skene is more tied to a particular platform, i.e, EJB,

but could be applied to any other problem domain by

merging suitable domain information.

3 UNIFIED PERFORMANCE

ENGINEERING FRAMEWORK

We propose a Model Driven Framework based on

various MDA(MDA, 2003) standards. Figure 1

depicts the overall UPE framework. All the

metamodel as well as transformations are based on

MOF(MOF, 2003). We claim that basing the design

model as well as analysis models on same

metamodel i.e. MOF will facilitate easy

transformation.

The main advantage of this approach is that

performance estimation can be done continuously

during the design stage. It will also maintain the

performance parameters separate from the design

model so as not to clutter the design model during

development. Another advantage will be that

transformation will be based on metamodel level

that will be expressed in a high-level model

transformation language compared to complex XML

based transformations.

We present the UPE approach in more details in

the following sub-sections by describing the various

models and the required transformations:

3.1 Design Model

As design evolves during the development process,

any of the UML features deemed fit by the

developers are to be used. Organizations usually

follow some process standards such as Unified

process, which prescribe the set of diagrams to be

used at different stages. However, in the initial stage,

we will focus on Use case diagram to represent the

workload, activity diagram to represent the

scenarios, and deployment diagram to represent the

resource allocations. In particular we will be using

UML 2.0 notations and diagrams as well as any

PSM specific profiles such as CORBA or J2EE.

3.2 Performance View

We propose a novel idea of defining performance

views to represent performance related input as well

as results based on the concept described in MOF

QVT (QVTP, 2003). This mechanism will allow any

other view, say, security model, to be described

without interfering with performance model on the

same design model. This will be an editable view

that will allow the selected UML diagrams from the

design model to be annotated with performance data

TRANSFORMATION OF UML DESIGN MODEL INTO PERFORMANCE MODEL - A Model-Driven Framework

577

by using an extension of UML profile for SPT(SPT,

2001).

3.3 Intermediate Performance

Model

From this performance view, the intermediate

performance model that combines performance

related information from all the design artifacts and

performance view is generated. This will involve a

transformation that will create the Intermediate

Performance model by systematically analyzing

each of the activity diagrams and combining their

information into one performance model. The IPM

metamodel is closely related to CSM approach of

(Petriu, 2004).

3.4 Performance Analysis Model

The transformation of this IPM will result in the

performance model of choice such as Queuing

Network Model, SPA, Stochastic Petri Nets, etc.

These different transformations need to be designed

to achieve the desired performance model. This

model then will generate the parameters that will be

submitted to the Performance tool for solving the

model.

3.5 Transformations

Various model transformations will be needed for

converting Design Model annotated with

performance into Intermediate Performance model

and for transforming Intermediate performance

Model into Performance Analysis Model.

An editable view generation is required for

performance annotations to be specified on the

software design model.

3.6 Feedback

The results from the performance tools will be sent

back and represented in the performance view. This

will also need similar reverse transformations which

we are not currently focusing upon.

As QVT based approach will lead to

standardization in the model transformation field, we

are planning to use one of the suggested approaches

for our research work. We will be using OCL 2.0

(OCL, 2003) , MTL from QVT Partners (Biju, 2003)

that is based on the relational approach (Akehurst,

2003) Although some of the required specification

standards are not completed by the standards body,

we will apply the concepts, not from the point of

view of complying with the standards but in a way

to validate them for a specific domain such as

Design Model

(UCD, AD,

Performance

View

Intermediate

Performance

Model

MOF

UML

metamodel

Perf View

metamodel

IPM

Metamodel

Performance

Metamodel

Performance

Model

InstanceOf

View

Transformation

Figure 1: Unified Performance Engineering Framework.

ICEIS 2006 - INFORMATION SYSTEMS ANALYSIS AND SPECIFICATION

578

performance. We are hopeful that this may

contribute to the field of model driven development

by identifying shortcomings and proposing

alternative solutions.

4 CONCLUSIONS

We have outlined UPE - a framework for model

driven software performance engineering in this

paper. The need for such an approach was

established; the relevant standards and technologies

that are under various stages of completion

themselves were described. Since the underlying

field of MDD itself is in a high state of flux, it is a

challenge to devise an approach that will be adaptive

to such changing scenario.

As model driven paradigm is increasingly being

embraced by researchers and industry, We are

hopeful that we can contribute to this pool of

knowledge by applying this approach to the domain

of Software Performance Engineering thereby

validating the same and making performance

analysis an integral part of the software development

activity.

REFERENCES

Akehurst D.H., Kent S., Patrascoiu O., 2003. A Relational

Approach to Defining and Implementing

Transformations between metamodels, Journal of

System and Software, Issue 2

Balsamo S., Di Marco A., Inverardi P., Simeoni M.,2004.

Model-Based Performance Prediction in Software

Development: A Survey. IEEE Transactions on

Software Engineering, vol. 30, n. 5, pp. 295-310

Appukuttan B., Clark T., Reddy S., Laurence T.,

Venkatesh R.,2003. A Model driven approach to

building implementable model transformations. In

Workshop on Model Driven Architecture: Foundations

and Applications, University of Twente, June 2003

D’Ambrogio A.,2005. A Model Transformation

Framework for the Automated Building of

Performance Models from UML Models. In

Proceeding of ACM Workshop on Software and

Performance.

Infrastructure, 2003. Object Management Group, Unified

Modelling Language(UML) Specification:

Infrastructure, Version 2.0, ptc/03-09-15, Sept. 2003.

MDA Guide, 2003. Object Management Group, MDA-

Guide, V1.0.1, omg/03-06-01, June 2003.

MOF, 2003. Meta Object Facility (MOF)2.0 Core

Specification, ptc/03-10-04, Oct. 2003.

OCL, 2003. Unified Modelling Language:Object

Constraint Language version 2.0, Draft

AdoptedSpecification, ptc/03-08-08, Aug. 2003.

Petriu D.., Woodside M, 2004. A Metamodel for

Generating Performance Models from UML Designs,

Proc. of UML Conference, LNCS 3273, pp 41-53 .

QVTP, 2003. QVT-Partners. MOF

Query/Views/Transformations, Revised Submission.

OMG Document: ad/2003-08-08

Skene J ,Emmerich W.,2003. Model driven performance

analysis of enterprise information systems, Electronic

Notes in Theoretical Computer Science(ENTCS),

Elsevier Science B.V.

SPT, 2001. UML profile for SPT, OMG Document:

ptc/04/02/01

Super, 2003. Object Management Group, UML 2.0

Superstructure,ptc/03-08-02, Aug. 2003.

WOSP, 98-05. Proceedings of ACM Workshop on

Software Performance, 1998, 2000, 2002, 2004, 2005

XMI, 2003. Object Management Group, XML Metadata

Interchange (XMI) Specification, version 2.0, May

2003.

TRANSFORMATION OF UML DESIGN MODEL INTO PERFORMANCE MODEL - A Model-Driven Framework

579