/m4

Primary LanguageClojureOtherNOASSERTION

What is MARATHON?

MARATHON (Modeling Army Rotation At Home or Not) is a mechanism for analyzing the effects of Army supply, demand, and policy variations, where supply is a set of potentially deployable units, demand is a set of activities requiring a unit, and policy is a collection of rules or constraints that determine a unit’s ability to fill a demand.

As a design goal, MARATHON seeks to validly simulate the physics of Army supply and demand, governed by policy, to analyze both the general behavior of such systems and the specific effects relative to changes in supply, demand, or policy. Ultimately, MARATHON is an analytic sandbox for evaluating courses of action relative to the Army Force Generation domain.

Army Force Generation

Army Force Generation is a system for managing readiness, the ability for units to deploy to meet contingencies. In general, force generation is the structured progression of increased unit readiness over time, resulting in the periodic availability of trained, ready and cohesive units prepared for operational deployment in support of civil authorities and combatant commander requirements. The domain of Army Force Generation is enormous, encompassing the range of processes and resources necessary to man, equip, train, deploy, and sustain the Army’s supply of units.

Out of necessity, MARATHON focuses on a subset of the Force Generation process, and generally holds many gross assumptions about the behavior of quite complex subsystems (such as training processes, manning, equipment, mobilization, etc.) Even with the Force Generation domain scoped to the unit level of detail , and with complex subsystems like equipping and manning abstracted away, the variety of supply, demand, and policy options is still staggering.

How Does MARATHON Work?

MARATHON typically simulates the force generation process through a coordinated set of supply, demand, and policy simulations.

Supply

The supply system acts as a coordination point for polling unit availability, a dissemination channel for simulation supply events, and a general container of units. Thousands of unique unit entities follow rotational policies that are either global (shared) or local (unique to the unit), and are directed by one or more supply systems to execute the “supply physics” dictated by the corresponding policy. Each unit’s simulated history can be traced, recorded, and reacted to within the simulation ecosystem.

Unit rotational policy generally consists of a directed sequence of states and durations. Units also have a behavior, which interprets policy to implement the desired supply-side and deployed actions. Policies are entirely modular and variable, as are individual unit behaviors. The decoupling of behavior and policy allows for both homogenous sets of units that appear to behave identically, as well as a diaspora of independent singletons that can apply similar behavior to different policies or interpret the same policies (via different behavior) to simulate radically different populations.

The potential for unique entities allows MARATHON to flexibly and modularly account for the legion of subtleties and corner-cases in the force generation problem domain.

Demand

Demands are activated, and slated for filling, based on a - potentially sophisticated - user-defined priority function. A fill system matches the highest priority demand to the most suitable supply as needed, and directs the transition of units from the supply system to deployments or other states. The fill system also accounts for potentially complex unit substitution rules, demand preferences, and almost any value function associated with the selection of units to fill demands.

Policy

Finally, a policy system accounts for changes to policy (such as ARFORGEN suspension, variation in lifecycle length, and changes in deployment time) by enacting system-wide policy changes in response to either time or event. Policy changes automatically filter down to subscribing units, enabling a rich and diverse simulation of the supply-policy-demand dynamics.

History

MARATHON was incepted at the Center for Army Analysis (CAA) on/around 2003, with various incarnations and ports over the years. Since 2010, Tom Spoon has served as the primary developer, maintainer, and architect for MARATHON. The current version of MARATHON, 4.x, is the simulation’s first port to Clojure and first entry into open source software.

Differences From Traditional Simulation Implementations

MARATHON 4 presents a significant departure from traditional Discrete Event Simulations, since the architecture embraces functional programming and persistent data structures wholly, rather than focusing on mutation.

MARATHON’s goal is to provide a simulation history, which is a sequence of simulation contexts - or snapshots - of the entire “world” in MARATHON. This history is lazily generated, yet allows us to define means for observing entities - typically the domain of clunky logging facilities and mutation in most simulations - that can operate on the entire stream of differential changes to the initial simulation context. MARATHON combines several novel features:

  • building the simulation transition functions (or step functions) as a composition of smaller functions
    • (as opposed to OOP-based classes or imperative mutation)
  • Optionally persistent, lazily computed stream of simulation history
  • a database layer based on the Component-Entity-System paradigm
  • Maintains the ability to have a classic observable/observer simulation model without sacrificing functional purity.
    • Where convenient (i.e. for logging, visualization, other side-effects).
  • Event-step (i.e. variable-width time-step) simulation.
  • High-level transforms on the simulation history.
    • Familiar idioms like map, filter, reduce work out of the box, since history is merely a time-indexed stream of simulation contexts.
    • Allows for precise reasoning about causality, tracing, debugging, etc.
  • Entity behaviors based on Behavior Trees.

Developer Notes

MARATHON 4.x is currently alpha software written to support research efforts. Consequently, it is undergoing constant verification and testing to ensure compliance with legacy models. Expect possibly significant change in the coming weeks/months prior to a formal release.

MARATHON 4.x also makes exclusive use of a supporting library called SPoRK (Spoon’s Operations Research Kit), among other open-source dependencies. SPoRk is a collection of utilities and other infrastructure (including ai, generic simulation, statistics, etc.) upon which MARATHON stands.

SPoRK is an independently originated, non-government open source project licensed under the EPL (to be released shortly).

Additionally, several sections of the code-base are “crusty”, and possibly annotated as obsolete due to porting from a different language, or failed experimentation.

There are ongoing efforts to prune the codebase of noise, and to enhance the documentation.

That being said, most of MARATHON is written in a somewhat literate style; that is, the source code may be processed by Marginalia to produce a companion set of documentation with inline prose.

Interested parties and code archeologists may visit the marathon.demo namespace to see high-level usage examples, or the marathon.ces.testing namespace to see soup-to-nuts tests of the simulation infrastructure.