ONR Special Notice N00014-21-S-SN11: Battlefield Communication Node with Weak state Implementation f
Abstract
"Approved for Public Release"Naval dominance in the 21st century relies heavily on the operational advantage gained from integrating, all domain networks and joint military assets into a unified any sensor-to-shooter connected warfighting system. Information sharin,g across multi-domain resources with diverse degrees of freedom - be it heterogeneous tactical data links, multi-modal sensors or un,manned long-endurance platforms - is critical to optimize decision making and maximize warfighting effects in delayed, disconnected,, intermittent and limited connectivity environment. Therefore, the future naval network deployment must ensure that critical informa,tion finds a path to the right user at the right location in highly contested and highly dynamic environment. In light of the above,, our proposed approach is intended to provide (1) a Software Defined Network (SDN) architecture for managing the naval network in a,way that is interoperable across different branches of the military and different domain networks, and (2) data-driven machine learn,ing methods for identifying the right degree of synchronization among distributed SDN controllers. A hierarchical deployment of SDN,controllers will be investigated that can balance the effects of centralized and distributed network control inside a mobile environ,ment. Unlike the commercial state-of-the-art SDN implementations (such as ONOS and OpenDaylight) that always synchronize every contr,oller and its network state with all other controllers, our approach instantiates a different state table and adapts the table size,to be maintained on each controller by considering practical resource constraints and dynamics as well as the diversity of applicati,on needs running in the network. In order to make efficient synchronization decisions, we propose DQ Scheduler, a novel method that,leverages deep reinforcement learning techniques. DQ Scheduler builds upon the state-of-the-art Deep Q Network (DQN) method. It deve,lops new mechanismsfor improving the robustness of learning in order to handle scenarios of network fragmentation and disconnected c,ontrollers. It also attempts to learn how exactly to compose the synchronization messages in order to carry the most important infor,mation needed by each controller, thereby reducing even more the bandwidth consumption. The proposed method will be capable of makin,g the right synchronization decisions even under uncertainty about the accurate network state viewed by each controller, referred as, weak (not necessarily accurate) state. The feasibility of the proposed robust and adaptive synchronization methods will be shown th,rough an emulation platform (Mininet) that interfaces with commercial SDN software (OpenFlow protocol and RYU controller). A demonst,ration will be developed to highlight the practical performance benefits we can achieve by adopting a learning approach for synchron,izing the SDN controllers.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 08, 2022
- Source ID
- N000142212147
Entities
People
- Leandros Tassiulas
Organizations
- Office of Naval Research
- United States Navy
- Yale University