Issues in Enhancing Simulation Model Reuse

1
Issues in Enhancing Simulation Model Reuse

• C. Michael Overstreet
• cmo_at_cs.odu.edu
• Richard E. Nance
• nance_at_vt.edu
• Osman Balci
• balci_at_vt.edu

2
Motivations for Simulation Model Reuse

• To reduce life-cycle costs
• model specification
• code specification implementation
• VV plans execution
• accreditation
• To reduce time until new simulation is available
• near instantaneous construction of new
  simulations
• To improve quality of new simulations
• based on trusted or efficient components

3
Perspective/terminology

• A simulation typically consists of
• A collection of interacting sim. models
• An infrastructure enabling interaction of those
  sim. models
• Mechanisms for observing or characterizing
  selected behaviors
• Mechanisms for user interaction with simulation

4
Fundamental assertions - 1

• Each simulation is constructed to meet a concrete
  set of objectives, such as
• Improve system performance
• planning, design
• Improve understanding
• scientific modeling managers intuition
• Reduce training time, improved quality
• correctness of some aspects may not be
  important
• Build a fun game
• laws of physics might be intentionally ignored
• Generate believable behaviors
• movies, background for training simulations
• Different objectives can imply different
  behaviors, correctness, accuracy, and performance
  requirements for the same object or situation.

5
Fundamental assertions - 2

• Objectives determine desired behaviors of models.
• Desired behaviors determine model content.
• Models are based on abstractions and assumptions.
• Appropriateness of abstractions depends on
  desired behaviors.
• The models used in simulations reflect sometimes
  subtle tradeoffs among speed, accuracy, included
  features, costs.

6
Thus

• Simulation model reuse must take both original
  and new objectives into consideration valid
  reuse requires consistency between the two sets
  of objectives.
• Similarly for model assumptions and constraints

7
Occams view of simulation

• The simplest, minimal model is best
• Ease of understanding
• Quicker implementation
• Reduced debugging
• Often most run-time efficient
• Improved reuse potential (perhaps)
• easier modification, if needed
• Bias towards elegant, simple
• Thus models should be just barely good enough to
  meet objectives.

8
Desirability of minimalist view?

• Does this enhance or impede reuse?
• Does this reflect an inappropriate 1950s view of
  computing
• Its a sin to waste a cycle.

9
Economic facts of simulation

• Costs are in development CPU cycles are free.
• Tyranny of better software and cheaper hardware
• User needs are often quite elastic if its not
  too expensive, its a requirement.
• Faster, cheaper hardware results in unanticipated
  new uses (e.g., real-time decision support)
• Many of todays cutting-edge simulations will be
  perceived as inadequate tomorrow.

10
Conflicting user needs

• Create total immersion interactive environment
• Create believable environment
• Create new simulations on demand
• Create simulations cheaply
• Incorrect behavior unacceptable
• Some incorrectness required
• Games
• Tutorials
• Execution efficiency vital

11
Example levels of reuse

• Plug n play no changes necessary
• ModSAF a successful example
• Existing model easily altered to provide new or
  modified behaviors
• Can result in significant cost benefit
• Modeling approach useful in new domain
• Reuse concepts, architecture, designs, etc.
• Infrastructure reused (e.g., HLA)

12
Impossible goal automated reuse of arbitrary
models?

• Page Opper showed that deciding if a collection
  of models meets a set of objectives is
  NP-complete.
• Overstreet Nance showed that deciding if two
  models are equivalent is unsolvable.

13
Feasible goal automated reuse of specially
constructed models

• ModSAF (OneSAF) can build new simulation by
  combining existing library of models as needed.
• Each model is built from consistent set of
  objectives so that it will interact with other
  models correctly.
• Adding a new model to library requires that it be
  built in conformance to these objectives.
• Still, a slight change in objectives can mean
  that reuse of these models is undesirable.

14
Key reuse issues research needed - 1

• Determining how to locate potentially reusable
  models
• Detecting incompatible objectives and assumptions
  among selected models
• Building models in such a way that reuse
  potential is enhanced
• Determining the level of granularity that best
  enhances reuse potential.

15
Key reuse issues research needed - 2

• Capturing and representing the objectives,
  constraints and assumptions of each model.
• Determining if constraints (such as speed,
  memory) will be met with selected collection of
  models.
• If individual models are valid, what does this
  imply about a new combination?

16
Comments on issues

• Some of these issues are well know to designers
  of Simulation Programming Languages, for example,
  granularity
• GPSS (and many current simulation programming
  languages) consists of a collection of reusable
  models, each easily parameterized.
• But building a new simulation is like writing a
  new program from scratch.
• Use of high level components results in faster
  development but loss of flexibility

17
No single solution

• Execution overhead
• Some models are run once and thrown away
• Some model executions must meet real-time
  deadlines
• Some are execution intensive but not real-time
• Some models need only be suggestive (wake of a
  ship at sea) others must be highly precise
  (fluid flow about a supersonic wing).
• A solution should be less expensive than the
  problem it solves
• we need both quick dirty simulations and
  well-documented, highly reusable simulations

18
Summary - 1

• Reuse is, in large part, motivated by economics.
• The changing costs of computing changes the
  models we choose to build.
• The changing costs of computing changes the
  economics of reuse
• Faster hardware makes execution inefficiencies
  due to reuse irrelevant

19
Summary - 2

• Key to reuse is the capturing of objectives,
  assumptions and constraints.
• Models can be designed for reuse, but it appears
  feasible only when objectives are well-understood
  and stable.
• Completely automated reuse appears scientifically
  infeasible.
• Automated support is more likely economical.