Federation Playback and Restart

The Missile Defense Agency seeks innovative technology for supporting (1) playback and (2) restart in parallel and distributed federations involving multiple simulations (federates) operating in (a) real time involving humans and/or hardware in the loop (e.g., training, testing, studying human factors, etc.) and/or (b) logical time to support as-fast- as-possible execution (e.g., analysis, acquisition, planning, decision support, wargaming, algorithm testing, etc.). Playback is normally used to visualize and/or analyze the results of a successfully executed federation, while restart is normally used to resume each of the simulations participating in the federation from a checkpoint file that was previously saved during a prior run. The WarpIV Kernel (currently being used by the MDASim program) already provides automated playback support for federations using its High-Performance Computing Run Time Infrastructure (HPC-RTI) and/or Local RTI Component Proxy (LrcProxy) Framework. Both of these High-Level Architecture (HLA) RTIs, along with the WarpIV Native Federation Object middleware services, are being used by (a) the Defense Intelligence Agency (DIA) Integrated Threat Analysis Simulation Environment (ITASE), and (b) the Advanced Framework for Simulation Integration and Modeling (AFSIM) that is being used by the Air Force, DARPA, and U.S. Space Force. The playback capabilities for logging/compressing data and supporting visualization/analysis/mining tools can also be extended to support other RTIs as well. Properly supporting federation-wide checkpoints and restarts is significantly harder than supporting playback, especially for simulations with very large code bases involving models with complex and fragmented dynamic memory structures (i.e., blocks of memory that point to other blocks of memory that point to other blocks of memory, etc.). For this reason, the bulk of the work outlined in this Phase I proposal focuses on the much harder goal of restarting federates from checkpoints operating within a federation. In preparation for writing this proposal, WarpIV Technologies invested several weeks of IR&D to develop, test, benchmark, and prove the feasibility of the novel approach outlined in this proposal, which was to overload the new and delete operators in C++ to automatically track memory segments, find embedded pointers, and adjust virtual function table (vtable) pointers for classes with virtual functions when restarting. These challenging goals to automate the serialization of classes through persistence was fully successful and accomplished with the added benefit of speeding up dynamic memory operations. This means that applications using the AutoPersistence capability outlined in this proposal will run faster, not slower. We believe that this approach can fully automate checkpoints and restarts for C++ federates operating in any federation without modifying federate models. Approved for Public Release | 23-MDA-11635 (9 Nov 23)