Content uploaded by Leandro Navarro
Author content
All content in this area was uploaded by Leandro Navarro
Content may be subject to copyright.
Assessing a Distributed Market Infrastructure
for Economics-Based Service Selection?
Rene Brunner1, Isaac Chao1, Pablo Chacin1, Felix Freitag1, Leandro Navarro1,
Oscar Ardaiz2,
Liviu Joitac3, and Omer F. Rana3
1Computer Architecture Department, Polytechnic University of Catalonia, Spain
{rbrunner, ichao, pchacin, felix, leandro}@ac.upc.edu
2Department of Mathematics and Informatics, Public University of Navarra, Spain
oscar.ardaiz@unavarra.es
3School of Computer Science and the Welsh eScience Centre, Cardiff University, UK
{l.joita, o.f.rana}@cs.cardiff.ac.uk
Abstract. Service selection is an important issue for market-oriented
Grid infrastructures. However, few results have been published on the
use and evaluation of market models in deployed prototypes, making it
difficult to assess their capabilities. In this paper we study the integration
of an extended version of Zero Intelligence Plus (ZIP) agents in a mid-
dleware for economics-based selection of Grid services. The advantages
of these agents compared to alternatives is their fairly simple messag-
ing protocol and negotiation strategy. By deploying the middleware on
several machines and running experiments we observed that services are
proportionally assigned to competing traders as should be in a fair mar-
ket. Furthermore, varying the environmental conditions we show that
the agents are able to respond to the varying environmental constraints
by adapting their market prices.
Key words: Automatic Resource Allocation, ZIP Agents, Decentralized
Economic Models, Service Oriented Grids
1 Introduction
Grid Computing leverages the power of thousands of resources distributed across
computers/supercomputers/clusters linked by networks (from Intranet to the In-
ternet). Through the concept of Virtual Organizations (VOs), the Grid enables
the dynamic composition of such resources into interoperable services, which
multiply exponentially the VOs added value. However, given the unpredictabil-
ity of the underlying platform (Internet), scalable realization of such synergies
(in both physical and organizational levels) poses serious challenges to modern
large scale distributed systems research. Contrarily to other distributed systems,
?This work was supported in part by the Ministry of Education and Science of Spain
under Contract TIN2006-5614-C03-01, and the European Union under Contract
CATNETS EU IST-FP6-003769.
2 Authors Suppressed Due to Excessive Length
Grids have many independent resource providers with varying access policies. In
addition to large sizes, the diversity of polices leads to a complex allocation task
that cannot be handled manually by users. Automatic and adaptive resource
management is the solution to these challenges.
The primary visions for Grid computing are utility computing infrastruc-
ture and Grid services/service providers. In utility computing, a third party
service provider hosts and manages the Grid solution dedicated to serving a
single organization or the needs of multiple ones. Customers only pay for the
used resources. Grid services/service provider’s modularity enables the dynamic
composition and coordination of e-services which can be exchanged or traded be-
tween Grid users or brokers, following the usage models from utility computing.
These two features enable for the automatic trading of Grid Services in Service
Oriented Grids (SOGs).
There has been recently an increase of interest in SOGs within the Grid com-
munity towards services that are often considered as a natural progression from
component based software development, and as a mean to integrate different
component development frameworks. A service in this context may be defined
as a behavior that is provided by a component to be used by any other com-
ponent. A service stresses interoperability and may be dynamically discovered
and used [1]. Utility computing assumes service instances are created on the fly
and automatically bound as applications that are configured dynamically. The
service viewpoint abstracts the infrastructure level of an application. It enables
the efficient usage of Grid resources and facilitates utility computing, especially
when redundant services can be used to achieve fault tolerance. A SOG system is
configured on-demand and flexibly, which means different components are bound
to each other late in the composition process. Thus, the configuration can change
dynamically as needed and without loss of correctness. Decentralized Grid Mar-
kets based on agents have been proposed as suitable coordination mechanisms
for Grids and SOAs [2]. The market here is nothing more than a communication
bus it is not a central entity of its own and does not participate in matching
participants requirements using some optimization mechanisms. Direct agent to
agent bargaining allows participants to use the negotiation strategy more suit-
able to its objectives and current circumstances. Local bilateral bargaining also
facilitates the scalability of the system and the quick adaptation to fluctuations
in resource allocation dynamics. This enables for high scalability in both physical
and organizational levels. These concepts have been capitalized in [3] for SOG
purposes.
In this paper, we address the design and economic behavior of an economic
SOG infrastructure and its evaluation based on the Grid Market Middleware
(GMM), a resource allocation middleware which incorporates decentralized eco-
nomic models [4]. We show an economic adaptability of the agents that incorpo-
rate a decentralized economic algorithm based on an extension of the ZIP Agent
[5]. The experiments prove a seamless integration of the economic components.
Furthermore, economic reaction to environmental change in the resources type or
the demand is shown. The proposed economic matchmaking mechanism has the
Title Suppressed Due to Excessive Length 3
advantage to scale easily to large and dynamic environments, keeping the ability
to provide important information and shouted prices from other agents, which
rest incertitude or an approximation in alternative decentralized approaches.
2 Related Work
Economy based resource allocation has received a great deal of attention in the
last years. The GridBus Project [6] is a reference in SOG and utility based com-
puting, and has proposed a great variety of market models and tools for the
trading of Grid Resources. However, its strong emphasis on computational in-
tensive Grids and the hierarchical nature of some of the proposed components,
like the Grid Market Directory, diverges from the fully decentralized resource al-
location mechanism proposed here. Centralized approaches exist such as [7], but
scalability issues both in size and computational requirements further compli-
cate its applicability to large size Grids. Tycoon [8] is a market-based system for
managing compute resources in distributed clusters or Grids. It uses distributed
auctions with users having a limited amount of credits. Users who provide re-
sources can, in turn, spend their earnings to use resources later.
A few papers address fully decentralized market mechanisms for computa-
tional resources. In [9], a peer-to-peer (P2P) double auctioning mechanism is
proposed which builds on Zero-Intelligence agents (ZIP Agents) [5]. It was shown
that the results with original ZIP agents in continuous double auctions (CDAs)
depend strongly on the availability of the complete set of bid and offers coming
from all buyers and sellers, and the commitment to winner-to-winner allocations.
But a P2P or fully decentralized trading mechanism must be free of any central
authority for scalability reasons. To oppose these problems Ogston [10] propose
a P2P agent auction with centralized clusters. This offers complete information
about other traders in the same cluster, it assures price stability and it copes
as well with scalability issues of distributed systems. Another fully decentralized
approach is the one adopted with the catallactic agents [11]. In this approach
bilateral negotiations are established between a set of learning agents, and the
spontaneous coordination arises from both the bargaining and co-evolutionary
learning processes.
However, none of these approaches provide the infrastructure for integrating
explicitly the market based algorithms into service oriented Grids. Tycoon has
been used mostly in a clusters environment, and GridBus is provided as com-
plete software toolkit, not as a service. Our approach is to offer the economic
algorithms as Web Services for a seamless integration in any SOG.
3 Service Oriented Grid Market Middleware
3.1 The Grid Market Middleware
The Grid Market Middleware (GMM) provides the mechanisms to register, man-
age, locate and negotiate for services and resources. It allows trading agents to
4 Authors Suppressed Due to Excessive Length
meet each other based on its requirements and engage in negotiations. Further-
more, the middleware offers a set of generic negotiation mechanisms, on which
specialized strategies and policies can be dynamically plugged in. The GMM has
a layered architecture (Figure 1), which allows a clear separation of platform spe-
cific concerns from the economic mechanisms, to cope with highly heterogeneous
environments. A detailed description of both the design and implementation of
the GMM architecture can be found in [4].
Applications interact with the GMM in order to obtain the Grid services
required to fulfill the application tasks. The Base Platform supports the appli-
cation by providing a hosting environment for the Grid services. When a client
issues a request, the application determines which grid services are required to
fulfill it. These grid services represent either software services (e.g., a data pro-
cessing algorithm) or computational resources. The application service translates
these requirements to a WS-Agreement format [12] which is submitted to the
Grid Market Middleware. The middleware searches among the available service
providers, which have registered their particular service specifications, like con-
tractual conditions, policies and QoS levels. When a suitable service provider
is found, the application requirements re negotiated within the middleware by
agents who act in behalf of the service providers as sellers and the application as
buyers. Once an agreement is reached between the trading agents, a grid service
instance is created for the application. Afterwards a reference is returned to the
application, which can invoke it.
Fig. 1. Layered middleware architecture
3.2 The extended Zero Intelligence Economic Agents
In this work we consider a simplified Grid market with only one homogeneous
Data Mining service being traded. The execution time of the service can be
varied during the experiments. The auction mechanism is a continuous double
auction in which the agents follow a modified ZIP strategy based on [5].
In the context of the GMM, the buyer agents are called ComplexServices
(CSs) and the seller agents BasicServices (BSs). CSs aggregate BSs from the
Title Suppressed Due to Excessive Length 5
market. As BSs and CSs get involved in trading, the price will evolve by the
offer and the demand, with dependence on the limited CS budget and the limited
resources which can be sold by the BS. Once the BS has sold its resource to a
CS, it cannot accept more bids from other CSs CFPs until the moment when
the client of the awarded CS ends the execution of the sold Data Mining service
in the resource.
For the realization of the decentralized continuous double auction we divide
the traders in subgroups, called bidding clusters (see Figure 2) which are trading
independently. This allows to cope with the scalability of large networks an.
Moreover this approach enables the agents to be well-informed of shouts from
other agents, which in decentralized auctions is a general problem [9]. To avoid
that groups are only trading isolated, agents have to join and leave the clusters.
The selection of individual agents to move to another cluster depends on their
trading success. This method allows reaching one global equilibrium price P0for
all clusters situated the distributed market place. As the feasibility of a global
P0is already shown in [10], we will concentrate our prototype analysis focusing
in one bidding cluster.
B
S
S
S
S
B
B
B
A
Bidding Cluster
Bidding
Cluster
Bidding
Cluster
Bidding
Cluster
Bidding
Cluster
Bidding
Cluster
Fig. 2. Bidding clusters containing sellers S, buyers B and an auctioneer A.
Each cluster deploys an own central continuous double auction. The agents
are coordinated in a synchronous manner and are acting in bidding rounds.
Therefore a delegated auctioneer controls the matching of the bids and offers;
6 Authors Suppressed Due to Excessive Length
the highest bid corresponds to the lowest offer. No matching of a trade will be
executed, if no offer exists lower than the highest bid.
3.3 Interface with Application
In a SOG infrastructure, the GMM is exposed to be accessed by applications
trough a convenient access point, a Web Service which can be deployed in any
application server and integrated as a service in an existent SOG. Figure 3 de-
scribes the main steps in the interaction trough the access point. When a client
issues a request, the application determines which Grid services are required to
fulfill it. These Grid services represent either software services (e.g., a data pro-
cessing algorithm) or computational resources. The application translates these
requirements into a standardized WS-Agreement [12]. The application invokes
the access point and passes the corresponding WS-Agreement request. This is
in turn parsed and processed at the access point, which instantiates the GMM
with the required economic agents to fulfill the client request.
Fig. 3. Service Oriented Grid (SOG) infrastructure.
The GMM searches among the available service providers, which have regis-
tered their particular service specifications, like contractual conditions, policies
and QoS levels. When a suitable service provider is found, the application re-
quirements are negotiated within the middleware by agents who act in behalf
of the service providers as sellers and the application as buyers. Once an agree-
ment is reached among the trading agents, a Grid service instance is created for
the application and a reference is returned to the application/client, which can
invoke it.
Title Suppressed Due to Excessive Length 7
The server-side infrastructure is deployed by a set of scripts which allow
for the bootstrapping of BSs in available resources. The scripts perform the
automatic deployment and configuration of the BSs, which are then ready to be
contacted by CSs. Services offered by BSs for clients executions are also deployed
and exposed in Apache Tomcat application servers. Complemented by the access
point, this comprises a complete infrastructure for economic-based SOGs.
4 Prototype Application
4.1 Data Mining Grid Services Application
Different types of applications can be constructed and benefit from using the
GMM in the Grid, such as enabling the creation of VOs for planning, scheduling,
and coordination phases within specific projects or businesses. The ability of a
free-market economy to adjudicate and satisfy the needs of VOs, in terms of
services and resources, represent an important feature that markets can provide.
Such VOs could require a large amount of resources which can be obtained from
computing systems connected over simple communication infrastructures such
as the Internet.
As a proof of concept of the system model, we provide an application of
the GMM with extended ZIP agents to an existing decentralized free-market
prototype, the Catallactic Data Mining application [13]. The basic problem ad-
dressed by the data mining process is one of mapping low-level data (which are
typically too voluminous to understand) into other forms that might be more
compact (for example, a short report), more abstract (for example, a descriptive
approximation or model of the process that generated the data), or more useful
(for example, a predictive model for estimating the value of future cases). At
the core of the process is the application of specific data-mining methods for
pattern discovery and extraction. This process is often structured into a discov-
ery pipeline/workflow, involving access, integration and analysis of data from
disparate sources, and to use data patterns and models generated through in-
termediate stages. Selection and conversion of datasets as well as the execution
of the data-mining algorithm itself are the typical required steps. In the Catal-
lactic Data Mining services prototype, two Data Mining Services encapsulating
data conversion and algorithm execution are combined in a workflow achieving
a solution to the overall problem. For simplicity we restrict to the deployment of
the core Data-Mining Service, and we consider the pre-processing step as given
by the application.
Consider a scenario where a client issues sequential requests for Data Mining
services. The CSs try to map the incoming workflows to an available set of
services. The BSs, try to sell their services to the CSs which are instantiated
after successful negotiation upon the client request. Figure 4 shows a scenario
with two service types in the service market.
8 Authors Suppressed Due to Excessive Length
Fig. 4. Prototype Application using the SOG infrastructure.
4.2 Deployment and Experimental Setup
The bidding algorithm is based on extended ZIP agents. This allows reaching the
equilibrium price P0, at which the maximum resources will be exchanged, with
simple agents. Therefore they have to know the minimum price of the shouted
offers, by sellers Smin and the maximum price of the shouted bids by buyers
Bmax. These two values build the basis for the agent’s bidding algorithm to
calculate its new price P(t+1).
Algorithm 1: Bidding algorithm of the BS (seller).
Input:random1>0 and <0.2;
Input:random2>0 and <0.2 and not random1;
if Smin > Bmax then
Ptarget =Smin -(random1*P(t)+random2);
else
Ptarget =Bmax +(random1*P(t)+random2);
endif
priceChange = γ* priceChange + (1-γ) * β* (Ptarget -P(t));
P(t+1) = maximum (P(t)+ priceChange, Pmin) ;
Title Suppressed Due to Excessive Length 9
Algorithm 2: Bidding algorithm of the CS (buyer).
Input:random1>0 and <0.2;
Input:random2>0 and <0.2 and not random1;
if Smin > Bmax then
Ptarget =Bmax +(random1*P(t)+random2);
else
Ptarget =Smin -(random1*P(t)+random2);
endif
priceChange = γ* priceChange + (1-γ) * β* (Ptarget -P(t));
P(t+1) = minimum (P(t)+ priceChange, budget) ;
We setup controlled experiments by deploying several instances of the GMM
in a Linux server farm. Each machine has a 2 CPU Intel Xeon at 2.80 GH
and 2 GB of memory. The nodes in the farm are connected by an internal
Ethernet network at 100 Mps. The topology is a mesh: all interconnected. CFPs
are transmitted via groupcast to all the nodes in the destination groups (in our
scenario CFPS are groupcasted from CSs to BSs).
We deploy the GMM in 8 nodes. Four nodes host a BS each and the Data
Mining Web Service and other four nodes host the CSs, access points and clients.
The Web Services are exposed in Tomcat servers. Access for execution of these
Web Services on the resource node is what is traded between BSs and CSs. The
experiments consist in launching 4 clients concurrently, which use each one of
the CS as broker. Each client makes requests to the CS and leaves the market
after a successful trade. It will re-enter a proceeding round with the probability
of 1
3. Whenever a CS wins a bid with a BS, it invokes the Data Mining Service
in the selected node, and the resource in the corresponding node gets locked for
the duration of the service execution. We measure the selling prices of the BSs
and observe the proportion of successful CFPs issued by the CSs.
5 Experiments and Evaluation
The goal of the experiments is to show the performance of the GMM as an auto-
mated economic-aware resource management tool by means of the Data Mining
Grid prototype application. The extended ZIP agents are expected to show an
effective and fair trading, which can be measured with the price and the allo-
cation rate of each agent. Varying the technical parameters of the environment,
we expect price adaptation of the agents in the marketplace.
5.1 Idealized Experiments with Idle Resources
The experiments are sensitive to a competitive use of other processes, because
this might cause an increase of the Data Mining WS execution times. Therefore
we make first experiments with idle resource, which warranties stability of Data-
Mining Services execution times.
10 Authors Suppressed Due to Excessive Length
50
60
70
80
90
100
0 200 400 600 800 1000
price
time(s)
price
(a) resource execution time of 3000 ms
50
60
70
80
90
100
0 200 400 600 800 1000
price
time(s)
price
(b) resource execution time of 100 ms
Fig. 5. Price evolution with varying offer with and a constant demand rate of 1
2
Besides the effect of changing the offer, also the variation of the demand for
the resources needs to be proved. Therefore we change the probability that a
CS reenter the market (by issuing a new demand) after a successful trade. The
comparison between Figure 4 and Figure 5 shows clearly the significance of the
demand. In Figure 4 the demand rate probability of re-entering the market is 1
6,
which keeps the amount of the CS low and decreases the price. Figure 5 show
the price decrease when the CS re-enter the market after every successful trade
(probability of 1).
50
60
70
80
90
100
0 200 400 600 800 1000
price
time(s)
price
(a) demand rate 1
6
50
60
70
80
90
100
0 200 400 600 800 1000
price
time(s)
price
(b) demand rate of 1
Fig. 6. Price evolution with varying demand rate with and a constant execu-
tionTime of 1000 ms.
Title Suppressed Due to Excessive Length 11
5.2 Adaptation to Different Constrains
The experiment in this section illustrates the adaptation of the prototype for a
changing set-up environment. In this section the execution the time of the Data-
Mining Services is varied to obtain real scenarios where processed input data-sets
sizes might differ. To simulate such cases, the execution time of the resources
will vary during the running time of the experiment. It changes iteratively, every
200 seconds the executions time from high (3000 ms) to very low (100 ms).
50
60
70
80
90
100
0 200 400 600 800 1000 1200
price
time(s)
price
Fig. 7. Varying task load (WS execution time) dynamically. phase 0 (t = 0 s -
450 s): stabilization;phase 1 (t = 450 s - 650 s): WSexecTime = 100 ms; phase
2(t = 650 s - 850 s): WSexecTime = 3000 ms; phase 3 (t = 850 s -1050 s):
WSexecTime = 100 ms; phase 4 (t = 1050 s -1200 s): WSexecTime = 3000 ms.
After a stabilization phase of about 450 seconds (phase 0), the experiment
in Figure 2, shows price adaptation to varying market constrains in form of task
loads (the WS execution times). From a short resource execution time (like 100
milliseconds) results that the market contains many offers. Consequently the
prices of the product decreases. Contrarily, decrementing the offer by setting the
execution time to 3000 milliseconds leads to an increasing price.
5.3 Process Competition
Increasing the realism of the environment, we consider an experiment were the
nodes in the cluster run with other competing processes which influence the
resource performance. This has an impact on offers of ressources which should
12 Authors Suppressed Due to Excessive Length
be considered by the agents. We show how agents effectively react to the process
competition by adapting prices.
50
60
70
80
90
100
0 200 400 600 800 1000
price
time(s)
price BS-76
active BS-76
match BS-76
(a) BS
50
60
70
80
90
100
0 200 400 600 800 1000
price
time(s)
price CS-71
active CS-71
match CS-71
(b) CS
Fig. 8. Prices with competing process.
The allocation rate in Figure 9 shows the distribution of over 4000 matched
trades. A nearly equal distribution of the resources to the CS can be seen as
well as the nearly equal distribution of the bought BS resources can be seen.
Even in a real application with uncontrolled process competition an almost fair
allocation is obtained.
0
10
20
30
40
50
60
70
80
75 76 77 78 79 80
percentage
agent
allocation rate
(a) Allocation rate of BS
0
10
20
30
40
50
60
70
80
70 71 72 73 74 75
percentage
agent
allocation rate
(b) Allocation rate of CS
Fig. 9. Allocation rates in a experiment of competing process.
5.4 Evaluation
The results of the three experiments demonstrate how a simple decentralized
economic algorithm based on ZIP can be plugged into the GMM infrastructure in
Title Suppressed Due to Excessive Length 13
order to allocate resources to client in service oriented applications, by achieving
automatic and fair trading of resources between Grid clients and Grid service
providers, mediated respectively by the CS and BS agents.
Furthermore the results show that the agents react to changes in the eco-
nomic environment. The accepted price reflects the variations in demand (trough
demand rate) and offer (trough varying execution time of the services, which re-
sults in varying resource availability). It can be seen that the price increases
when the demand also increases (Figure 6) and that correspondingly the price
increases when offer decreases (Figure 5), as a result of more time consumption
by services. Nevertheless the distribution of allocations between buyers and sell-
ers remain proportioned (Figure 9), as expected in fair markets. Consequently
it follows that the prices will increase in case of large-scale failures or delays.
Moreover, this automatic price correction behavior is able to react to dynamic
conditions in underlying Grid resources (Figure 7).
6 Conclusions
We have shown a complete infrastructure for economic-based SOGs and we have
demonstrated its application in a Data Mining SOG prototype. The proposed
infrastructure provides both the scripts for automatic bootstrapping of traded
Grid services and the agents selling the services at the Grid service provider
side, as well as the Web Service access point for the seamless usage of the SOG
infrastructure by clients. The economic agents (employed an extension of ZIP
agents) are able to operate in a decentralized environment, automatically evolv-
ing trading prices with varying offer and demand rates. This has important
advantages in large scale Grids over computational costly centralized solutions,
mainly scalability and feasibility in open and decentralized systems.
The experimental results in Section 5 show that agent-based trading of re-
sources at stable prices can be achieved using the GMM. Moreover the allocation
of traded resources is well-balanced among the seller agents as well as among
the buyer agents. Our analysis demonstrates that the agents are stable against
economic changes in their environment. The agents overcome dynamics of the
system by conserving the expected offer and demand reactions in fair market.
Future work comprises the inclusion of more complex workflows and its in-
tegration/evaluation in the architecture, increasing the size of the test bed, and
test the infrastructure with additional prototype applications. The need for more
decoupling of individual agent behavior configuration from middleware will lead
eventually to the design of an independent Economic Agents Framework plug-
gable as a service in the current infrastructure.
References
1. I. Foster, C. Kesselman, J. M. Nick, and S. Tuecke, “Grid services for distributed
system integration,” Computer, vol. 35, no. 6, pp. 37–46, 2002.
14 Authors Suppressed Due to Excessive Length
2. T. Eymann, M. Reinicke, W. Streitberger, O. Rana, L. Joita, D. Neumann,
B. Schnizler, D. Veit, O. Ardaiz, P. Chacin, I. Chao, F. Freitag, L. Navarro,
M. Catalano, M. Gallegati, G. Giulioni, R. C. Schiaffino, and F. Zini, “Catallaxy-
based grid markets,” vol. 1, (Amsterdam, The Netherlands, The Netherlands),
pp. 297–307, IOS Press, 2005.
3. T. Eymann, M. Reinicke, F. Freitag, L. Navarro, O. Ardaiz, and P. Artigas, “A
hayekian self-organization approach to service allocation in computing systems.,”
Advanced Engineering Informatics, vol. 19, no. 3, pp. 223–233, 2005.
4. O. Ardaiz, P. Chacin, I. Chao, F. Freitag, and L. Navarro, “An architecture for
incorporating decentralized economic models in application layer networks,” Mul-
tiagent Grid Syst., vol. 1, no. 4, pp. 287–295, 2005.
5. C. Preist and M. van Tol, “Adaptive agents in a persistent shout double auction,”
in ICE ’98: Proceedings of the first international conference on Information and
computation economies, (New York, NY, USA), pp. 11–18, ACM Press, 1998.
6. R. Buyya, D. Abramson, and S. Venugopal, “The grid economy,” in Proceedings of
the IEEE, vol. 93, pp. 698–714, 2005.
7. B. Schnizler, D. Neumann, D. Veit, and C. Weinhardt, “Trading grid services -
a multi-attribute combinatorial approach,” European Journal of Operational Re-
search, in Press, 2006.
8. K. Lai, B. A. Huberman, and L. Fine, “Tycoon: A distributed market-based re-
source allocation system,” tech. rep., HP:arXiv:cs.DC/0404013, 2004.
9. Z. Despotovic, J.-C. Usunier, and K. Aberer, “Towards peer-to-peer double auc-
tioning,” in HICSS ’04: Proceedings of the Proceedings of the 37th Annual Hawaii
International Conference on System Sciences (HICSS’04) - Track 9, (Washington,
DC, USA), p. 90289.1, IEEE Computer Society, 2004.
10. E. Ogston and S. Vassiliadis, “A peer-to-peer agent auction,” in AAMAS ’02:
Proceedings of the first international joint conference on Autonomous agents and
multiagent systems, (New York, NY, USA), pp. 151–159, ACM Press, 2002.
11. T. Eymann, B. Padovan, and D. Schoder, “The catallaxy as a new paradigm for the
design of information systems,” in Proceedings of the 16th IFIP World Computer
Congress, Conference on Intelligent Information Processing, 2000.
12. WS-Agreement, “Web services agreement specification.”
https://forge.gridforum.org/sf/projects/graap-wg, 2007.
13. L. Joita, O. F. Rana, F. Freitag, I. Chao, P. Chacin, L. Navarro, and O. Ardaiz, “A
catallactic market for data mining services,” Future Gener. Comput. Syst., vol. 23,
no. 1, pp. 146–153, 2007.
