The FCD advances stand for the custom end-to-end orchestration of fully distributed and programmable networking assets, as a pool of software and hardware components available for control and data plane functions. Detaining multi-objective characteristics, the optimization aspects involved on a fluid network include a myriad of resource parameters and capabilities (e.g., cost, availability, performance). In its foundation, the fluidity depends on tailoring such optimization aspects on-demand to attend service guarantees as much as infrastructure operational capabilities. By definition, this approach encompasses network slicing, horizontally and vertically spanning technological/administrative domains and multiple protocol stacks, respectively. From central clouds, to central office sites, to mobile base station sites, and even into the mobile terminals themselves, i.e. one fluid “cloud” from device to datacenter, hosting telco functions as well as third-party applications that require adequate, service-tailored connectivity. Altogether, this STA aggregates scientific and technological advances via the following enlisted key research topics:
- Application-tailored Network Architectures: To meet applications needs advanced network programmability and automation must support different QoS-sensitive services (e.g., XR, robotics, gaming), therefore demanding an end-to-end flexible network architecture. New protocols, establishing cross-layer communication and fine-tuned by orchestration mechanisms take place in the research outcomes that might shape future networking architectures, bringing even more close applications needs into the network [Lachos et al., 2020].
- Network Slicing 2.0: the effective realization of end-to-end multi-domain network slicing poses major challenges on the realization of 5G network slices [Foukas et al., 2017] to natively embrace clouds. The support of such a continuum, from border to cloud, requires new mechanisms that blur the boundaries of current network slicing. New shapes of slicing need to be provided for services beyond 5G, including the aggregation and decomposition of the existing slice models, a vision recently outlined in [Galis et al., 2020].
- Networking as Systems: advanced networking programmable languages have become the status quo together with enhanced programmability of network functions. Jointly, they can establish mechanisms that offload data plane and control plane functions on-demand, improving packet processing metrics. The decomposition of network functions brought by cloud networking models concern the core of this approach. Still, more effort must exist on the development of programmable interfaces and automation mechanisms to deliver networking control and data planes as flexible and adaptable system of systems.
- In-Network Computing: recently, “the network is the computer” paradigm has been identified [Sapio et al., 2017] as a complementary powerful mechanism to the traditional end-system computing model. New networking programmable hardware and languages [Bifulco and Rétvári, 2018] like P4 are turning this approach into reality. One exemplar application is In-band Network Telemetry (INT) based on P4 data plane functionality that allows fine-grained metric computation at each hop and processed at sink nodes. While in its infancy, proven useful in-network computing needs further work from multiple disciplines (HW, SW, etc.). Relevant related efforts are being devoted at IRTF Computing in the Network Research Group (COINRG).
- Intent-Based Networking (IBN): intent has become a method to specify a high-level policy for networking behavior, expressing “what” must be accomplished instead of the “how” [Clemm et al., 2020]. IBN is envisioned to facilitate the automation and distribution of networking tasks for cutting-edge applications. For its realization, closed-loop automation must be in place inside the operation and management of networks. Different mechanisms that establish the definition of intent resolution, and its included mechanisms sit still in need of further maturity to become concrete in production networks. Examples of such efforts include the definition of intents, frameworks to realize intent interpretation (e.g., using natural language processing), extensive automation of orchestration control-loops, and monitoring and actuation intent workflows detaining reliable resiliency boundaries.
- Time-Sensitive Networking (TSN) / Deterministic Networking (DetNet): reliable margins of networking metrics sit in the core of TSN and DetNet [Nasrallah et al., 2019], specifically for the industry 4.0 and audio/video applications. Mostly, such applications require strict networking behavior, e.g., zero congestion loss. Already in place in standardization models (IETF DetNet WG and IEEE 802.1 Task Group), TSN and DetNet establish operational requirements for networks that must be guaranteed by fine-granular monitoring and control signaling mechanisms. Such efforts must be clearly elaborated for different use cases and together with orchestration critical control loops, fastly reacting in run-time to meet the stringent service requirements of latency and jitter [Finn and Thubert, 2019].