A Distributed Publish/Subscribe System for RDF Data - Data Management in Cloud, Grid and P2P Systems - page 42

discussion is given about how duplicates are avoided when in-memory buffers
overflow. Moreover, subscriptions are formulated through an ad-hoc scripting
language and no experimental evaluation is available.

3E entC oudD gn

In this section we give a description of the data and subscription model used by
our system, dubbed EventCloud. We explain how events and more specifically
how RDF data, along with subscriptions, are indexed in a CAN network.

3.1 Data and Subscription Model

Our data and subscription model follows the approach taken in [15] to allow users
to formulate queries and subscriptions but also to insert and publish information
with the same models, that is respectively RDF and SPARQL.

Events. The data are expressed in the RDF model using 4-tuples (quadruples)
whose elements are named RDF terms. In our system an RDF term may be
either an IRI or a Literal value. Elements generated at the same time by a given
source form a Compound Event ( CE ), as defined by (1b). Each CE is made
of a list of quadruples and all quadruples share a common term called graph
value. This term is built with a combination of a unique source identifier and
a timestamp. The purpose of this graph value is twofold. It is used to identify
the event source, the event itself and also to offer the possibility to link together
several quadruples for emulating, yet unbounded, multi-attribute values like in
traditional pub/sub systems.

q =( g,s,p,o ) | g,s,p,o ∈ RDFTerm (1a)

CE =( q 1 , ..., q i , ..., q n ) | q i =( g,s i ,p i ,o i ) (1b)

The EventCloud is based on a four dimensional CAN overlay that uses the lexi-
cographic order for routing requests. The four dimensions of the CAN coordinate
space are mapped respectively to the graph, the subject, the predicate and the
object of any RDF 4-tuple that is indexed. One benefit of this approach is that
a quadruple represents a point in the four dimensional Cartesian space. Hence
a quadruple will only be stored by a single peer of the overlay. This indexing
approach has several advantages. First, it supports range queries (looking for
values in a specified range) eciently. Second, the lexicographic order preserves
the data semantics so that is gives a form of clustering of quadruples sharing a
common prefix. In contrast, hash-based approaches destroy the natural ordering
of information and make the management of complex queries dicult and ex-
pensive. The Figure 1 shows how CEs and subscriptions are mapped to a CAN
network.