Evaluating a Bio-Inspired Approach for the Design of a Grid Information System - Advancing Artificial Intelligence Through Biological Process Applications - page 293

INTRODUCTION

Furthermore, a self-organization approach
based on ants' pheromone (Van Dyke & al., 2005)
enables each agent to regulate its activity, i.e. its
operation mode , only on the basis of local infor-
mation. Indeed, each agent initially works in the
copy mode: it can generate new descriptors and
disseminate them on the Grid. However, when it
realizes from its own past activity that a sufficient
number of replicas have been generated, it switches
to the move mode: it only moves descriptors from
one host to another without generating new rep-
licas. This switch is performed when the level of
a pheromone variable, which depends on agent's
activity, exceeds a given threshold.
The ARMAP protocol can effectively be used
to build a Grid information system in which (1)
resource descriptors are properly replicated and
(2) the overall entropy is reduced. A balance be-
tween these two features is achieved by regulat-
ing the pheromone threshold, i.e., by shortening
or extending the time interval in which agents
operate under the copy mode.
A semi-informed discovery protocol exploits
the logical resource organization achieved by
ARMAP. Indeed, whenever a large number of
descriptors of a specific class are accumulated
in a restricted region of the Grid, it becomes
convenient to drive query messages (issued by
users to locate descriptors of this class) towards
this region, in order to maximize the number of
discovered descriptors and minimize the response
time. While this chapter focuses on enhancements
and performance of the ARMAP protocol, the
discovery protocol, namely ARDIP ( Ant-Based
Resource Discovery Protocol ) is here shortly
discussed, whereas an extensive analysis can be
found in (Forestiero & al., 2007).
This chapter also introduces the SO-Grid ( Self
Organizing Grid ) portal, a Web portal which gives
remote users access to an event-based simulator
written in Java. This portal, available at the URL
http://so-grid.icar.cnr.it , allows for the experi-
mental reproduction of simulation results and can
be used by researchers to perform “parameter

To support the design and execution of complex
applications, modern distributed systems must
provide enhanced services such as the retrieval
and access to content, the creation and manage-
ment of content, and the placement of content at
appropriate locations. In a Grid, these services are
offered by a pillar component of Grid frameworks,
the information system . This chapter discusses
a novel approach for the construction of a Grid
information system which allows for an efficient
management and discovery of information. The
approach, proposed in (Forestiero et al., 2005) in
its basic version, exploits the features of (1) epi-
demic mechanisms tailored to the dissemination of
information in distributed systems (Peterson et al.,
1997, Eugster & al., 2004) and (2) self organizing
systems in which “swarm intelligence” emerges
from the behavior of a large number of agents
which interact with the environment (Bonabeau
& al., 1999, Dasgupta, 2004).
The proposed ARMAP protocol ( Ant-based
Replication and MApping Protocol ) disseminates
Grid resource descriptors (i.e., metadata docu-
ments) in a controlled way, by spatially mapping
these descriptors according to their semantic clas-
sification, so to achieve a logical reorganization of
resources. A resource descriptor can be composed
of a syntactical description of a Grid service (e.g.
a Web Services Description Language - WSDL
- document) and/or a semantic description of the
capabilities of the service.
Descriptor reorganization results from pick
and drop operations performed by a large number
of agents. Each ARMAP agent travels the Grid
through P2P interconnections among Grid hosts,
and uses simple probability functions to decide
whether or not to pick descriptors from or drop de-
scriptors into the current Grid host. This approach
is inspired by the activity of some species of ants
and termites that cluster and map items within
their environment (Bonabeau & al., 1999).