J. Vargas, R. F. DeMara, A. J. Gonzalez, M. Georgiopoulos, and H. Marshall, 
"PDU Bundling and Replication for Reduction of Distributed Simulation 
Communication Traffic," Journal of Defense Modeling and Simulation, Vol. 1, No. 
3, August, 2004, pp. 167 - 185.

Abstract:
Communication bandwidth and latency reduction techniques are developed for 
Distributed Interactive Simulation (DIS) protocols. DIS Protocol Data Unit (PDU) 
packets are bundled together prior to transmission based on PDU type, internal 
structure, and content over a sliding window of up to C adjacent transmission 
requests, for 1 < C < 64. At the receiving nodes, the packets are replicated as 
necessary to reconstruct the original packet stream. Bundling strategies 
including Always- Wait, Always-Send, Type-Only prediction, Type-Length 
prediction, and Type-Length-Time prediction are developed and then evaluated 
using both heuristic parameters and a gradient descent back-propagation neural 
network.
Several communication case studies from the One Semi-Automated Forces (OneSAF) 
Testbed Baseline (OTB) are assessed for multiple-platoon, company, and 
battalion-scale force-on-force vignettes consistent with Future Combat Systems 
(FCS) Operations and Organizations (O&O) scenarios. Traffic is modeled using the 
OMNeT++ discrete event simulator models and scripts developed for a hierarchical 
communication architecture consisting of eight enroute C-17 aircraft each 
carrying three Ethernet-connected M1A2 ground vehicles, a wireless flying LAN 
based on Joint Forces Command's Joint Enroute Mission Planning and Rehearsal 
System (JEMPRS) for Near-Term (JEMPRS-NT) and follow-on bandwidth capacities. 
The simulation traffic includes Opposing Force (OPFOR) control via a CONUS-based 
ground station via its corresponding satellite links. Different bandwidth 
capacities are simulated and analyzed PDU travel time and slack time, router and 
satellite queue length, and number of packet collisions are assessed at 64 Kbps, 
256 Kbps, 512 Kbps, and 1 Mbps capacities. Results indicate that a Type-Length 
prediction strategy is capable of reducing travel time up to 85%, slack time up 
to 97%, queue length up to 98% on bandwidth restricted channels of 64 Kbps.