Peer-to-peer (P2P) systems adopt a completely decentral-
ized approach for sharing data over the Web. In P2P sys-
tems [4] each user adds resources to the system. Usually,
peers have the same functional capabilities and responsibil-
ities and they do not depend on any centralized administra-
tion. Important issues in P2P systems are to choose the right
data placement approach across peers, and to ensure data
availability without incurring additional overheads. From
this point of view, P2P systems can be classi¯ed into two
main groups: unstructured and structured P2P systems [4].
Schema-based P2P systems have scalability problems of data
integration systems. These problems are solved by PDMSs
in combining P2P and distributed database techniques. Peers
join the system by providing their own schemas and by
matching them with other ones to discover their acquain-
tances for e®ective data sharing. In this paper, we consider
unstructured PMDMS and we make the following supposi-
tion and contribution: we focus on MFL to de¯ne multi-data
source schema and exchange queries between peers. MFL is
a simple and powerful language that allows users to de¯ne
the multi-data source schema and to formulate their need in
a single query. In PMDMS, each peer maintains a multi-data
source ¯le describing concrete data sources schemas and con-
°icts (in Con°icts data source) of semantically linked peers
(called neighboring peers). In fact, data sources present het-
erogeneity that consists of di®erences in names, types, etc.
Several perceptions of the same real world lead to di®erent
schemas. To integrate the data sources together, we ¯rstly need to solve con°icts between their schemas. The con°ict
data source, provided by multi-data source, owned by a peer,
is the basis of a semantic overlay network where peers hav-
ing similar elements constitute a semantic network neigh-
borhood. Con°icts data source are used further to address
queries routing (not addressed in this paper). In [15] we gave
a performance evaluation of queries routing based on multi-
data source approach with respect to important criteria such
as precision, recall, response time and number of messages.
Then we compared this result with the routing algorithm of
SenPeer [6], a PDMS developed recently with respect to a
hybrid (i.e. super-peer/peer) approach where peers are con-
nected to super-peers according to their semantic domains.
The remainder of this paper is organized as follows: section
2 gives related work concerning data integration. Section 3
describes our algorithm of semantic reconciliation between
peers and we conclude in section 4.