Many shortcomings affect today’s networks, and must be addressed if we expect successful 5G network deployments: overly long provisioning times; reliance on proprietary, hard-to-modify and cost-ineffective hardware devices and components; and the daunting complexity emerging from the need to converge a wide range of heterogeneous access technologies and multi-vendor network components.
SUPERFLUIDITY tackles these challenges with a multi-pronged comprehensive strategy:
- Flexibility, via an architectural decomposition of network components and network services into elementary, reusable primitives.
- Simplicity, via a cloud-based architecture, getting rid of access-specific gateways and integrating heterogeneous JBOAs (“just a bunch of accesses”) within a converged cloud-network infrastructure ranging from the core all the way to the network edge.
- Agility, via virtualization of radio and network processing tasks.
- Portability and viability, through platform-independent abstractions, permitting reuse of network functions across multiple heterogeneous hardware platforms, while allowing for the vendors’ need for closed platforms or closed source implementations.
- High performance beyond the state of the art, via software acceleration, specialization and adaptation to hardware accelerators, while making these mechanisms transparent to network service designers so that they can focus on the development of novel services and not performance optimization.
This high-level strategy will permit SUPERFLUIDITY to design and implement a novel, superfluid network architecture where network services can be deployed near-instantaneously, whenever and wherever they are needed, with high performance; this architecture will have the following properties:
- Location-independence: Network services can be deployed (or relocated) at various heterogeneous networks along the end-to-end path (at the operator, at the edge, next to the base stations or access points, at customers’ premises) following the performance needs of particular applications, and reducing latency by migrating processing as close as possible to the end customers.
- Time-independence: Deploying or moving processing tasks and virtualized contexts to a different location should ideally happen near instantaneously, in timescales in the order of tens of milliseconds, if the goal is to support intra-technology handovers and/or service continuity along with unnoticeable performance degradation to end customers.
- Scale-independence: Decouple network services from their scaling. Scale out shall become just an orthogonal property of a service, as fast and easy as booking additional resources and “sticking” them to it transparently.
- Hardware-independence: The entity developing or deploying the software-based processing should not have to worry about the details of the underlying hardware. This will result in taking transparently advantage of an huge range of heterogeneous hardware capabilities, from high-end x86 servers with virtualization extensions and FPGA boards all the way to low-power, low frequency ARM-based boards at the edge.
The above overall goal includes eight specific project objectives:
- Novel 5G data plane processing architecture – Design a flexible, open and programmable 5G data plane processing architecture and relevant APIs for network functions’ convergence
- Converged 5G platform – Design, implementation, and evaluation of a unified and high performance distributed cloud platform technology for radio and network functions support and migration.
- New Algorithms and functions – Design, development and evaluation of algorithmic and design improvements for radio processing tasks, flow processing primitives, and service optimization.
- Ultra-fast and efficient virtualization – Design, implementation and evaluation of beyond the state of the art quickly instantiable, low memory footprint, and high performance virtualization technology
- Hardware adaptation and abstraction – design and development of technologies and interfaces to exploit and integrate customized hardware.
- Control and provisioning framework – Extensions of existing and widespread frameworks for platform’s management, control, and elastic provisioning.
- Security framework – Security abstractions and mechanisms to control the access to, and execution of, the network processing functions, and to prevent third-party network functions from having a negative impact on other clients’ functions, the network, or the Internet at large
- Contribution to standardization – Feed SUPERFLUIDITY results into the relevant standards bodies and communities working on de-facto standard tools