Enhancing Coalition Networking using Software Defined Coalitions – An Overview
Military / Coalition Issue
Military missions require use of various infrastructure assets including communication links, computational servers, data storage, databases, sensors and other resources. Dynamicity and agility of military operations demand near real-time configuration, re-configuration and provisioning of these resources, while supporting efficient and robust sharing of assets among coalition partners or armed forces. State-of-the-art techniques currently cannot achieve this.
Core Idea and Key Achievements
To address the above issue, the DAIS ITA team has proposed a new architecture called Software Defined Coalitions (SDC), which significantly extends capabilities offered by the existing Software Defined Network (SDN). The key idea of SDN is to separate control functions from the communication switches and links, and implement the control functions by software on a centralized controller for each administrative domain of the network. Due to the software implementation of control functions, SDN is easy to change, flexible, reconfigurable, agile and efficient.
Besides communication resources, coalition operations also make use of other infrastructure resources such as data servers, storage, databases, sensors, etc. The proposed SDC architecture is composed of multiple domains of such resources, where various domains possibly owned by different coalition partners are dynamically joined together to form the SDC infrastructure. The main objective of the proposed SDC is to enable efficient, reconfigurable, agile and robust control and sharing of a variety of resources across domains and among coalition partners for supporting different military operations.
Communication switches and links form the Data Plane for transferring user data, while domain controllers are connected to form the Control Plane for exchanging control information. In addition to managing resources within each domain, the new technical challenges of SDC lie on the management and control of different types of resources across domains. Our design challenges include: (a) how, when and what information domain controllers synchronize with each other, (b) how the control plane can be made robust, (c) what architectural or programming abstractions should be used for exchange of resource status information among controllers, (d) how resources can be shared across domains, and (e) how SDC can quickly respond to infrastructure fragmentation and reconnection.
Key achievements on SDC include the development of designs and techniques for:
- SDC Controller Synchronization
- Resource Sharing in SDC to Support Coalition Missions
- Control Plane Architecture of SDC
- Robust Network and Learning Architectures for SDC
- Joint Reinforcement and Transfer Learning for Distributed Service Configuration in Fragmented SDC
In the process, the team has also developed novel techniques to overcome various machine-learning issues such as model complexity and excessive learning time, which are applicable beyond the control and management of SDC. The techniques include:
- Reinforcement Learning for Military Network Control
- Graph Attention Networks for Congestion and Mobility Prediction
- Binarized Neural Network
- State-Action Separable and Embedding for Reinforcement Learning
Implications for Defence
The collection of new techniques will enable defence to realise the concept of SDC for dynamic, agile and robust configuration, provisioning and sharing of infrastructure assets among coalition partners or armed forces, which are unmatched by our adversaries.
Readiness and Alternative Defence Uses
This work is technology readiness level (TRL) 2/3. Many of the SDC techniques have been prototyped in practical systems or environments, including the demos of controller placement and synchronization, resource sharing, robust network architecture, and learning in fragmented SDC. These techniques are ready for adoption, modification and enhancement for implementation on practical defence systems.
Resources and References
Please refer to the SDC Achievements and Demos as highlighted in this presentation for related publications and resources.
Imperial College, Yale University, IBM US, Purdue University, ARL, Dstl