
as the performance of the individual applications of
which it is composed. When we consider integrated
services in which EAI tools are used from the
viewpoint of performance, the time taken to
complete delivery of the overall integrated service
through workflow activity is one of the most
important performance measures. Traffic generation
by each application is governed by the workflow
process, so traffic patterns generated by applications
working together are correlated with each other.
Traffic from one application will often cause
multiple other applications to simultaneously
generate traffic. In such cases, an increase in traffic
from one application leads to an increase in traffic
from related applications. Therefore, in designing
the architecture and capacity of the network and
server resources, the IT-system designers will have
to cooperate with the application developers and
consider the workflow process and the
characteristics of application traffic invoked during
execution of the designed workflow process.
In this paper, we propose a methodology for
evaluating the performance of generic EAI systems
of the kind described above, i.e. where the broker-
server has a workflow engine that invokes
applications according to the workflow process. The
network is in a hub and spoke configuration with the
broker at the center. The proposed methodology
involves step-by-step modeling of the following
items: (i) the workflow process, (ii) the traffic-
communication paths of applications, (iii) the
network and server system, and (iv) defining the
applications invoked in the various phases of the
workflow activity (binding of applications to the
workflow process). For the proposed methodology
to be practicable, we developed OPNET (OPNET)
process and node models to realize the functions of
the above workflow-driven broker-server.
This paper is organized as follows. In the next
section, we briefly survey past works about the
performance engineering in application. In section 3,
we give an overview of the proposed methodology.
In section 4, we briefly cover OPNET modeling of
the workflow broker-server and the workflow
processing. In section 5, we describe a simple case
study of the application of this methodology. We
close with a very brief review and a couple of ideas
we are working on to improve the proposed
methodology.
2 PERFORMANCE
ENGINEERING IN
APPLICATION
The unified modeling language (UML) is currently
the best tool we have that encompasses the
information, business systems, and technical
architecture (Erikson, 2000). Korthaus proposed the
BOOSTER (Business-Object-Oriented Software
Technology for Enterprise Reengineering) process
(Korthaus, 1998; Schder, 1998) as a multilevel
approach to business-object-based system
development. A “multilevel process architecture”
defines the framework for the activities which have
to be performed. The BOOSTER process
architecture has four levels: business engineering,
system-architecture engineering, application
engineering, and “business-object component”
engineering. Korthaus has summarized the activities
in each level of engineering and the UML diagrams
that should be used in engineering activities
(Korthaus, 1998). Aoyama (2002) described another
framework for the creation of business-driven web
services. Aoyama’s model for the development of
web services consists of business-process,
application, and platform layers. The UML is the
common language for development of the
application software. When the model that defines
an IT system is written in UML, it is convenient to
use UML as the basis for defining models to be used
in evaluating the IT system. Pooley et al. (1999)
summarized past approaches to software-
performance engineering and proposed some ideas
on the exploitation of UML designs in performance
modeling. Here, the use-case diagram provides the
basis for the definition of workloads in the system.
Implementation diagrams provide the mapping onto
computing and storage devices. They are essential
to definition of contention and the quantification of
available resources. Correspondences are then
drawn between the implementation diagrams and an
underlying queuing model. Finally, the behaviour of
the above queuing model is obtained either through
queuing-network analysis or simulation. Balsamo et
al. (2003) proposed a simulation-based approach to
the performance modeling of software architectures
specified in UML. They defined a way to set up
process-oriented simulation models of UML
software specifications. Correspondences between
UML and simulator objects are as follows: use-case
diagrams correspond to the workload models, the
deployment diagrams correspond to resources
models, and the activity diagrams correspond to the
steps of the simulation. They called the prototype
version of this tool UML-Ψ, which is intended to
indicate “UML performance simulator.”
ICETE 2004 - GLOBAL COMMUNICATION INFORMATION SYSTEMS AND SERVICES
276