SEKE 2012 Proceedings - Knowledge Systems Institute

DETECTING EMERGENT BEHAVIOR IN
DISTRIBUTED SYSTEMS CAUSED BY
OVERGENERALIZATION
Seyedehmehrnaz Mireslami Mohammad Moshirpour Behrouz H. Far
Department of Electrical and Computer Engineering, University of Calgary, Canada
{smiresla,mmoshirp,far}@ucalgary.ca
Abstract— Distributed system design faces many challenges
due to lack of central control. Emergent behavior is a vital
problem in distributed systems and leads to unexpected
behaviors and major faults. Emergent behaviors are usually
categorized into two groups: emergent behaviors occur due to
scenarios incompleteness and emergent behaviors that are
produced as a result of synthesis of behavior models. In this
paper, the methods for detecting the second group of emergent
behaviors are studied and a technique for synthesis of behavior
models from scenarios is developed to be employed for
detecting emergent behaviors. The proposed technique detects
overgeneralization in the behavior model synthesis.
Overgeneralization happens as the result of behavior model
synthesis and depends on the assumptions of the process. In
addition, the proposed technique addresses the issue of the
existing ad-hoc methodologies by providing an automated
algorithm. This algorithm can be used by a syntax checker to
automatically detect the emergent behaviors in the scenarios.
The proposed algorithm is validated using a case study of a
mine sweeping robot.
Keywords; Distributed systems; Emergent behavior; Message
sequence chart; Overgeneralization.
I. INTRODUCTION
A distributed system is composed of several components
without a central controller [1]. Distributed systems
specification is usually composed of a set of scenarios that
show system components and messages sent between them.
Scenarios are generally shown using Message Sequence
Charts (MSCs) or sequence diagrams [2, 3]. Behavior of the
components can be generated using the scenarios.
A popular approach for analyzing the behavior of system
components, is going from scenarios to state machines, i.e.
synthesis of behavior models [4-6]. Each component is
modeled by a state machine and the messages that are sent
and received determine the states and transitions.
In the process of behavior model synthesis, several
behaviors not originally intended for system may happen
that are called em ergent behaviors [8-10]. They are eith er
unexpected situations that may be caused by incompleteness
of the scenarios or produced as the results of behavior model
synthesis. Detecting emergent behavior in early design
stages reduces the deployment costs significantly [11].
There are two types of emergent behaviors: first group is
detected by comparing the behavior models and scenario
specifications by considering the properties of the system.
The second group is result of overgeneralization [7]. Hence,
it depends on the rules and assumptions of the behavior
model synthesis.
Only a few works have targeted the overgeneralization
problem. In [7], synthesis of state machines from scenarios
is proposed. However, the relationships between scenarios
are rather ambiguously defined. Blending scenarios is
necessary as it provides a comprehensive overview of
system behavior. Two methods are proposed to combine the
scenarios: state identification and scenario composition. In
state identification [5, 6, 12], the components are f irst
modeled with different states in the state machines and
similar component states a re identified and combined to
enable the scenarios to merge. In [2], another approach for
merging scenarios is proposed where scenarios are split to
smaller parts with lower complexity and high-level MSC
graphs are used to blen d the smaller sequence of behavior
since they are simpler to manage.
In [13], a technique to identify the identical states that
may cause emergent behaviors is proposed and a method to
synthesize emergent behaviors by merging identical states is
presented. This leads to detecting emergent behavior due to
presence of identical states. However, although identical
states are the potential causes of emergent behavior, not all
identical states may lead to e mergent behavior. The
shortcomings of this technique are: first, all identical sta tes
are used for merging the state m achines, and
overgeneralization is inevitable; and second, it relies on
designer to decide which states causes emergent behavior.
As ad-hoc methods are unreliable and inefficient, in this
paper, we replace th e ad-hoc methodology in [13] by an
automated method. A technique is devised to distinguish the
identical states that cause emergent behaviors. To have a fair
comparison with [13], the same mine sweeping robot system
used in [13] is considered.
70