It is well known that the system performance of parallel discrete event simulations depends on the assignment of workload to processors. If one processor is heavily loaded while the others are lightly loaded or idle, the overall performance of the system may be improved by offloading some of the workload to a less loaded processor. Load migration is the means by which this workload is dynamically moved from one processor to another. Central to the successful implementation of load migration is the ability to effectively predict processor loading and to apply domain relevant heuristics to select the source, the entities, and their destination for migration.

The modified discrete-event simulation (DES) scheduling paradigm used by battlefield simulations impedes the system's ability to effectively predict processor loading. In this paradigm, the real-time or scaled real-time systems depend on the busy wait construct to cycle through a time-delay loop and fire an event at the correct time. Because these cycles are consumed during what would otherwise be idle time, the inefficiencies inherent in this type of polling are moot with respect to processing time at the application level. However, at the system level, where processor loading is normally ascertained, since all of the cycles are fully consumed by the application, there is no practical way of predicting the processor loading.

In this paper we will present a modification to the DES scheduler used in real-time systems such that it allows for the identification of trends in processor loading. Briefly, this modification will monitor the queue access request so it is possible to generate and analyze the pertinent data. Next, the paper will discuss how items are selected for offloading once the processor has been determined to be in the "overloaded" state. This involves implementing a number of application and system configuration heuristics as they relate to the entity-based simulation domain. Finally, the paper will present our sample implementation and the results of our analyses.