adfa, p. 1, 2011.
© Springer-Verlag Berlin Heidelberg 2011
Global Pricing in Large Scale Computational Markets
Lilia Chourou
1
, Ahmed Elleuch
2
and Mohamed Jemni
1
1
LaTICE laboratory, Higher School of Sciences and Techniques of Tunis, University of Tunis,
Tunis, Tunisia
lilia.gherir@planet.tn, Mohamed.jemni@fst.rnu.tn
2
CRISTAL laboratory, National School of Computer Sciences, University of Manouba,
Manouba, Tunisia
Ahmed.Elleuch@ensi.rnu.tn
Abstract. Scale up the number of computing resources is a challenging issue
when building a global computational system. For this purpose, we present an
approach that adopts the commodity market model as an economic incentive
model and ensures the balance between supply and demand. We show how this
model may be adapted and applied to a large scale computational infrastructure.
To achieve a competitive equilibrium, prices are adjusted according to a tâton-
nement like process. However, this process, like several other pricing algo-
rithms proposed in the literature, does not fulfill the scalability requirement: all
prices of all commodities are computed by only one auctioneer. In the present
work, a fully distributed pricing algorithm is proposed based on an existing par-
tially distributed version. While in this last version, the price of each commod-
ity is computed by only one auctioneer, in our algorithm, a variable number of
auctioneers is used. To each auctioneer is associated a limited number of con-
sumers and suppliers with low communication delay. Our algorithm is then
scalable with respect to the number of suppliers and consumers. To evaluate our
algorithm, we have performed a simulation study. For different number of auc-
tioneers per commodity, the experimental results show that our algorithm con-
verges as well as the partially distributed version. Moreover, by splitting the
search space among auctioneers, our algorithm accelerates the convergence.
Keywords: pricing, competitive equilibrium, scalability, global computational
system.
1 Introduction
One major problem facing high-throughput applications is the need for large-scale
computational resources. Global computing systems are a cost-effective alternative to
parallel computers. They are able to provide a great number of idle resources donated
by volunteers and accessible via the Internet. However, the number of available re-
sources still needs to be expanded. For this purpose, we have to apply an economic
incentive model that would allow increasing the number of suppliers and consumers
by embedding a strategy for welfare. The economic model would also be used to re-
solve the load balancing problem. In this paper, we have adopted the commodity mar-