Application Framework for Programmable Network Control - Advances in Network-Embedded Management and Applications - page 39

approach is to implement the network functions as part of the application, i.e.
create and manipulate a virtualized network (overlay network). If the network
has limited resources to accommodate application demands in a best-effort
manner, frameworks exist to manage the quality of service on behalf of the
application [12-14]. Next, we illustrate some approaches from related network
research areas that deal with these limitations.
Overlay networks enable developers to redesign and implement, amongst
others, addressing, routing and multicast services optimal to their application
domain [15]. Overlay networks are widely used to support specific services,
such as distributed hash tables [16], anonymity [17], and message passing
[18]. Overlay networks might lead to sub-optimal utilization of network
resources, because the mapping to the physical network resources is not open
to the application developer. Moreover, overlay networks essentially duplicate
functions offered by the physical network. Recently, some efforts [19]
propose to expose physical network properties to applications to improve their
mapping to the physical network. Assuming that networks are properly
dimensioned, at least from the user's perspective, overlay networks are a
straightforward solution to support their specific network service

requirements.
Sensor networks illustrate best limitations in network resources. Sensor
networks motivate tight integration of applications and network services [20].
Because of the resource constraints, sensor network designers attempt to use
the scarce resources efficiently and various approaches to program sensor
networks have been developed [21]. In macroprogramming [22], high-level
programs use an intermediate language to abstract away concurrency and
communication aspects in sensor application programming. A compiler
translates the programs into basic instructions for individual nodes, and takes
communication characteristics into account. In TinyDB [23], communication
is integrated with a data query mechanism. Macroprogramming and TinyDB
show that with a framework that structures the design space of network
control applications, it becomes possible to design and implement reusable
components for new applications.
Our research in advanced applications of networks [24-30] shows that
applications have different optimal network services. Existing network
management systems do offer APIs to configure network services [31]. Such
APIs implement the network abstractions chosen by the network operator. We
found that our use cases in hybrid networks and sensor networks require more
flexible and specific network services than those designed and implemented
by network operators. Because the application domain offers developers more
flexibility, it might be more practical to implement network services as part of
the application. Hence, we developed a model that enables developers to
program networks as part of their application [32]. The resulting framework,
User Programmable Virtualized Networks (UPVN), models the interworking
between networks and applications and provides a conceptual framework to