Peer-to-Peer Overlay Network Based on Swarm

Intelligence

Vesna Sesum-Cavic and Eva Kühn

Vienna University of Technology, Institute of Computer Languages

Space Based Computing Group, Argentinierstr. 4, 1040 Wien, Austria
{vesna,eva}@complang.tuwien.ac.at

Abstract. As the number of information in the Internet constantly increases and

the complexity of systems rapidly grows, locating and manipulating complex
data has become a difficult task. We propose a self-organizing approach that
combines purely decentralized unstructured peer-to-peer (P2P) with space
based computing in order to effectively locate and retrieve information from a
network. The approach is inspired by swarm intelligence, is distributive and
autonomous. As the scalability is a common open issue both for unstructured
P2P networks and for coordination models, our approach successfully copes
with that by using a biologically inspired multi-agent system. Benchmarks
demonstrate powerful query capabilities with a good scalability.

Keywords: swarm intelligence, unstructured P2P, space based computing,

lookup mechanisms.

1 Introduction
The highly dynamic nature of the Internet, characterized by a huge number of infor-
mation often expressed as complex data, imposes the necessity for new and advanced
ways for locating and retrieving of information. We propose a self-organizing archi-
tecture as a combination of unstructured P2P [1] and space based computing ([4], [9])
for searching and retrieving data concurrently. It intends to use the benefits of both
paradigms. The lookup mechanism is inspired by swarm intelligence ([6], [8], [12]).

2 Architecture and Design

In our approach, a space-based architecture, called XVSM (extensible virtual shared
1
memory) , based on a formal model [7], is used . It generalizes Linda tuple space
communication ([4], [9]) by introducing customizable, shared data structures (shared
containers [10]). A container stores entries that generalize tuples. Each container
possesses so-called coordinators (e.g. fifo, random, template matching, key, label),
and can be located and accessed via a URL. All operations on it (read, take, write)
use this URL to refer to the container. When a container is created, it can be made
publicly accessible by a name. [2] combines a structured P2P system with XVSM. In
contrast, in this paper, we propose the creation of an overlay network that consists of
1
Project and open source download site: www.xvsm.org

H. Aldewereld, V. Dignum, and G. Picard (Eds.): ESAW 2009, LNAI 5881, pp. 65–67, 2009.
© Springer-Verlag Berlin Heidelberg 2009
66 V. Sesum-Cavic and E. Kühn

lookup containers. A container can be published under one or more public names.
The discovery of data is performed by means of swarm intelligence. For the struc-
ture of the overlay network, we use the scale-free network approach [3] for an initial
construction. Software agents realize the algorithms of the overlay network acting in
swarms and performing the role of artificial ants. The multi-agent system is inspired
by ant colony that is distributed, self-organizing, with a high level of autonomy.
Writing of information. An ant can put the content (i) randomly (no need for swarm
intelligence algorithms), or (ii) using brood sorting [5]: entries are distributed on the
basis of their type (similar entries stay closer to each other). In this case, the similarity
function δ is based on spatial locality. Retrieving of information. We implemented
two different ant algorithms, an adaptation of MMAS that is combined with Local
Search (2.5-Opt) and an adaptation of AntNet. We introduced the following changes:
In the ConstructSolution procedure [8], heuristic values from the random proportional
rule are interpreted as a quality of the used links, expressed in time needed to traverse
a particular path from lookup container A to lookup container B by using a particular
link. The possible resulting situations of ants’ search for the particular data are: no
data found, exact data found, and acceptable data found with the accuracy/error rate
< ε, where ε is a parameter given in advance, connected to the definition of δ. The
DepositPheromone procedure [8] is changed as follows: If an ant on its trip *) found
exact data, it deposits pheromone; *) found acceptable data, it deposits less amount of
pheromone, *) did not find data, then skip depositing pheromones on its trip (i.e., the
values on arcs it traversed will be the same as the values on the rest of unvisited arcs
in the network). String comparison (representing URLs) uses a similarity function δ,
based on a spatial locality. A different amount of pheromones is deposited according
to the quality of solution found. The general form δ is: δ = δ (currentSolution, exact-
Solution), that describes how good (acceptable) the found solution is, δ ∈ [0,1]. In
case of changing the type of δ, its value can be scaled into the same segment [0,1].
DepositPheromone procedure is changed: 1) for MMAS algorithm: Δτ =1/MCbest
where M=1/δ; 2) for AntNet algorithm: τ := r⋅(1-τ)⋅δ.

3 Simulation Results
We implemented two ways respectively of writing data into containers and of perform-
ing lookup and retrieval of data from containers on a cluster of 4 computers (2*Quad
AMD 2,0 GHz, 16 GB RAM). A first group of benchmarks investigates the behavior of
the algorithms in different combinations (Table 1) and different parameter settings [8]
while trying to find the best possible combination of parameters. When comparing the
two kinds of lookup, AntNet supplies the best performance in case of retrieving only
one query (Fig. 1). A second group of benchmarks investigates the query capability of
our system and compares it with Gnutella’s lookup mechanism [1]. We can conclude
that the adapted intelligent algorithms cope successfully with an increasing number of
queries, compounded of several simple queries. We used the possibility of increasing
the number of concurrently working ants. Our approach outperformed the Gnutella
lookup by means of the obtained performance. Fig. 1 represents one test case using 80
containers.
 Peer-to-Peer Overlay Network Based on Swarm Intelligence 67

Table 1. Possible combinations

MMAS AntNet
Random case 1 case 2
Brood sorting case 3 case 4

Fig. 1. Different lookup mechanisms

4 Conclusion
Locating and retrieving of data in a global network implies the necessity for effective
lookup mechanisms. We introduced a self-organizing approach combining decen-
tralized unstructured P2P with space based computing, that focuses on the quality of
the solutions and the time needed to obtain that solution, applies a learning principle
from the nature, and outperforms Gnutella lookup. Future work will consider large
size benchmarks, and other intelligent approaches, like bee intelligence [11].

References
1. Androutsellis-Theotokis, S., Spinellis, D.: A survey of peer-to-peer content distribution
technologies. ACM Computer Survey 36, 335–371 (2004)
2. Bessler, S., Fischer, A., Kühn, E., Mordinyi, R., Tomic, S.: Using Tuple-Spaces to manage
the Storage and Dissemination of Spatial-temporal Content. Journal of Computer and Sys-
tem Sciences (to appear, 2009)
3. Caldarelli, G.: Scale-Free Networks. Oxford University Press, Oxford (2007)
4. Carriero, N., Gelernter, D.: Linda in Context. CACM 32(4), 444–458 (1989)
5. Casadei, M., Menezes, R., Viroli, M., Tolksdorf, R.: A Self-organizing Approach to Tuple
Distribution in Large-Scale Tuple-Space Systems. In: Hutchison, D., Katz, R.H. (eds.)
IWSOS 2007. LNCS, vol. 4725, pp. 146–160. Springer, Heidelberg (2007)
6. Di Caro, G., Dorigo, M.: AntNet: Distributed Stigmergetic Control for Communications
Networks. JAIR 9, 317–365 (1998)
7. Craß, S., Kühn, E., Salzer, G.: Algebraic Foundation of a Data Model for an Extensible
Space-Based Collaboration Protocol. In: IDEAS, Calabria, Italy (to appear, 2009)
8. Dorigo, M., Stützle, T.: Ant Colony Optimization. MIT Press, Cambridge (2005)
9. Kühn, E.: Virtual Shared Memory for Distributed Architectures. Nova Science Publ. (2001)
10. Kühn, E., Mordinyi, R., Keszthelyi, L., Schreiber, C.: Introducing the Concept of Custom-
izable Structured Spaces for Agent Coordination in the Production Automation Domain.
In: AAMAS, Budapest, Hungary (2009)
11. Sesum-Cavic, V., Kühn, E.: Instantiation of a Generic Model for Load Balancing with In-
telligent Algorithms. In: Hummel, K.A., Sterbenz, J.P.G. (eds.) IWSOS 2008. LNCS,
vol. 5343, pp. 311–317. Springer, Heidelberg (2008)
12. Tolksdorf, R., Menezes, R.: Using Swarm Intelligence in Linda Systems. In: Omicini, A.,
Petta, P., Pitt, J. (eds.) ESAW 2003. LNCS (LNAI), vol. 3071. Springer, Heidelberg
(2004)