Page 68 - Kaleidoscope Academic Conference Proceedings 2022

2022 ITU Kaleidoscope Academic Conference

a significant challenge to networks with highly dynamic 6.1 Network configuration and adaptability
topologies. The constantly changing data plane connectivity
of satellite networks [8] constitutes a representative example, Self-driving capabilities in large network infrastructures,
where the use of IP protocol is rather restrictive since the with quick detection and reaction to network events, can be
relation between network nodes can change as well as that achieved by the decentralization of network intelligence.
between the nodes and the user endpoint. Another example This, however, will require a communication protocol (and
is that of vehicular networks, where the communication can relevant east-west interfaces) to synchronize the local views
be over direct links between vehicles and roadside of the distributed decision-making entities and to coordinate
infrastructure but where it can also involve multi-hop their actions so as to avoid inconsistent configurations (e.g.
connections between vehicles [11]. As in the case of satellite the introduction of loops). At the same time, the reduced
networks, the dynamics of communicating nodes result in computational capability of management nodes, as a result
fluid topologies, which require maintaining and updating of decentralization, will make it difficult to cope with
topological IP addresses on a frequent basis. collecting and processing the vast amount of data generated
in the network. Hence, a key challenge concerns the
The address structure could instead be flexible enough to development of lightweight telemetry mechanisms that
support multiple semantics that apply to environments with execute close to the data source and can strike the right
highly dynamic topologies. In addition to reducing the balance between accuracy and overhead. Such a mechanism
complexity of such networks, a flexible addressing scheme could use, for example, machine learning-based
can allow for richer policies that enhance packet treatment in classification methods for automatically reducing the
terms of routing performance and security. A representative dimensionality of data by efficiently discarding non-useful
example is the approach in [26], which uses semantic information.
addresses to represent virtual switches at fixed space
locations (based on geo-coordinates) being traversed by Concerning the programmability aspect, intent-based
satellites, and which can result in reduced service disruptions networking has been gaining traction in the last few years
caused by satellite handover events. evidenced by efforts in standardization bodies [6], [15], the
emergence of relevant workshops, and the inclusion of the
Table 1 – SoTA and challenges summary topic in international conferences. Besides the need of a
common intent specification language, key challenges
Area State-of-the- Research Challenges include a generalized intent decomposition mechanism, the
Art incorporation of feedback to ensure the continuous
Network Long-lived Automation: enforcement of intent, the automated selection of the most
configuration configurations, decentralized appropriate action(s) to execute (given multiple options) for
and adaptability centralization, management, light- achieving a specific objective, and tools to ensure
OpenFlow, P4 weight telemetry configuration consistency.
High-level
programmability: intent 6.2 Cloud-native networking
decomposition,
configuration The embedding of computation in the network poses several
consistency challenges which mainly stem from the requirement that
Cloud-native Communication Efficient resource resource management algorithms should no longer treat
networking and management algorithms
computation (joint optimization), different resource types in isolation but instead optimize
treated multi-provider resource them jointly. In conjunction with the need of real-time
separately federation reconfigurations to support demanding applications
Network SDN, NFV Network operating executing in the network, such algorithms should be
softwarization system, abstractions, designed with efficiency as a prime objective. In addition,
APIs, common the resource scarcity in edge computing environments
functionalities naturally forces the use of container-based technologies,
Network Fixed address Elastic addressing which operate on a finer granularity compared to
addressing length and scheme, semantically- conventional virtualization technologies and can achieve
semantics enhanced addresses and better usage. The management logic for coordinating the
routing mechanisms container-hosting locations, the allocation of user requests
among the distributed set of application execution points,
6. CHALLENGES AND RESEARCH and the decisions concerning the usage of resources will need
OPPORTUNITIES to follow a multidimensional optimization framework.
This section discusses the main research challenges In edge compute scenarios it is likely that multiple providers
associated with the four areas described in the previous parts will need to collaborate to form a large cloud infrastructure
of the paper that can allow for more flexibility in the network. both in terms of resources and geographical footprint, so as
These are summarized in Table 1. to support a wide range of services and large customer bases.
To this end, another challenge not only concerns the design

– 22 –