against subscriptions according to their temporal correspon-
dence rather than on content only. We introduce the concept
of temporal utility that is used for intelligent buffer manage-
ment and for content routing. Depending on the interest
of subscribers and their periodicity in the area of an infos-
tation, only publications which will be relevant in the next
future, as predicted by the subscription profile of the infos-
tation, are routed onto this infostation. This mechanism is
also used for buffer management on infostations with buffer
constraints, such as maybe mobile carriers.
Recent work on Delay Tolerant Networks (DTN) [4, 1]
has focused on unicast routing [8, 10] and on opportunistic
communication [14]. To our knowledge the only other work
concentrating on multicast routing and temporal issues for
delay tolerant networking is [19], where an approach to mul-
ticast is introduced trying to account for temporal group
membership. Previous work in this area has been based on
the exploitation of epidemic-style techniques [17, 11].
Some approaches which tackle location based dissemina-
tion with temporal constraints in specific scenarios like sen-
sor networks from a pure algorithm perspective have been
recently presented [7]. However, these approaches mainly
focus on the problem of tuning the replication of messages
in order to reach a certain group of nodes, given some spatio-
temporal constraints, rather than on supporting decoupled
communication with temporal semantics using a publish/-
subscribe architecture. In [6], Huang et al. investigate the
problem of delivering messages to a large set of nodes in a
manner that satisfies a potentially dynamic set of spatio-
temporal constraints without any infrastructure.
The solution that we are presenting in this paper is the
first attempt of designing a content-based routing mecha-
nism that allows for an intelligent buffer management and
forwarding of content based on the evaluation of the temporal
validity and constraints of subscriptions and publications.
The paper is organised as follows: Section 2 introduces a
number of scenarios to motivate our work further. Section 3
describes our approach in detail and Section 4 our evalua-
tion of the performance through simulation. In Section 5 we
compare our solution with the state of the art, while Sec-
tion 6 concludes the paper illustrating possible future work.
2. APPLICATION SCENARIOS
While developing our approach we have kept in mind a
number of scenarios, such as:
Advertising: Let us consider the case of a person regularly
travelling downtown to reach her workplace during week
days. She may register an interest in receiving informa-
tion to her mobile device about both travel news on certain
routes around her specific travelling times and of restaurant
promotions for lunchtime. Her subscriptions, then, will be
periodic ones, around 8am and 6pm for travelling on certain
trains and around lunch time for lunch promotions (only
Mondays to Fridays).1 Infostations around town, with quite
limited/slow connectivity to the backbone network2, will
register her presence in particular locations in periods of the
day and of the week: this will allow the temporal profiling
1Some of these subscriptions may happen implicitly by the
system registering a mobility pattern and linking that to
some sort of content interest.
2An example may be infostations that communicate period-
ically with a central server using a GSM network.
of the infostation in terms of which kind of subscribers are
in the area at the different times of the day or of the week:
this information is useful for both the selective download
of the content from the backbone (given the slow/limited
connectivity) and the buffer management, if this is limited.
In general, we envisage a scenario where infostations are
in charge of delivering the information to the subscribers
passing by using short-range wireless technologies such as
Bluetooth or 802.11. The business case is an interesting
one: service providers could sell specific time slots in specific
places to different advertising/travel companies according to
the content interests of the users seen in the past in those
places at those times. The billing of the service may be
based on the number of the publications actually delivered
to the users.
Remote Areas: Let us consider a remote village connected
to a bigger town through a bus route. Information can be
carried through the busses (with the help of mobile infos-
tations installed on the busses) between the town/Internet
and the village. A number of papers have focussed on this
model such as DakNet [12] and the system developed by
the University of Waterloo for delay tolerant communica-
tion support in India [13]. Information could for instance
be related to stock values or market prices for goods. As
the memory on the infostation is reasonably limited and
the network link to it is quite slow, it makes sense to have
techniques to selectively store only the information which is
going to be relevant in the village both in terms of content
and on timing.
3. TACO-DTN
In this section we illustrate our approach. We first define
the general topology of the systems we consider, then we il-
lustrate the algorithmic details of TACO-DTN. Our purpose
is to enhance the typical pattern matching mechanism used
in content-based routing with a notion of time-awareness
that improves delivery in scenarios characterised by tempo-
ral constraints and delay tolerant traffic.
3.1 System Topology
We assume a system composed of fixed infostations, mo-
bile (subscriber) nodes and, in some cases, mobile infosta-
tions which act as carriers (e.g., infostations on busses).
• Fixed infostations are connected (in some cases inter-
mittently or with slow links) to a backbone and pos-
sibly to the Internet. Content is sent to infostations
and stored temporarily in their (potentially limited)
buffer.
• Mobile subscriber nodes can be PDAs, mobile phones,
laptops, embedded devices for instance in cars, and
the like. The nodes are intermittently connected to the
infostations, as described in the scenarios in Section 13.
• Mobile infostations are infostations which move around
(e.g., busses) and can potentially carry information
from fixed infostations to subscribers in a different
(potentially remote) location. They will often have
limited buffers and intermittent, or slow, connectivity.
3We assume no multi hop routing between the mobile nodes:
routing is only considered for forwarding among infostations
or the Internet and the infostations.
84