million entity distributed simulations
Download
Skip this Video
Download Presentation
Million Entity Distributed Simulations

Loading in 2 Seconds...

play fullscreen
1 / 10

Million Entity Distributed Simulations - PowerPoint PPT Presentation


  • 98 Views
  • Uploaded on

Million Entity Distributed Simulations. Improving network load by eliminating broadcast interest groups. Presented by: David Prody. State of the Art. In 2005 the Institute for Defense Analysis conducted a million plus entity simulation.

loader
I am the owner, or an agent authorized to act on behalf of the owner, of the copyrighted work described.
capcha
Download Presentation

PowerPoint Slideshow about ' Million Entity Distributed Simulations' - paloma


An Image/Link below is provided (as is) to download presentation

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.


- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -
Presentation Transcript
million entity distributed simulations

Million Entity Distributed Simulations

Improving network load by eliminating broadcast interest groups

Presented by:

David Prody

state of the art
State of the Art
  • In 2005 the Institute for Defense Analysis conducted a million plus entity simulation.
  • Joint Semi Autonomous Force (JSAF) was used as the driver for entities inside the simulation
  • Scalable Parallel Processor (SPP) where use to provide the horsepower for the simulation
  • These nodes where spread out across the country including several supercomputer sites
    • Maui High Performance Computing Center
    • Aeronautical Systems Center Major Shared Resource Center
    • High Performance Computing Modernization Program (DOD)
how they did it
How they did it
  • Tree of Meshes Routing
    • Groups of nodes used mesh overlays to communicate with other local nodes
    • One node in each local mesh was connected to backbone mesh connecting the 5 primary simulation sites
  • Interest declaration
    • Modified Runtime Infrastructure version s (RTI-s)
      • Simulation nodes specify what data they are interested in.
      • Publish Subscribe model
    • Sender side quelch
      • Original RTI-s used multicast to handle interest groups
      • Smarter routers not only send out data to nodes that have requested it
  • Limited “active entities”
    • Most entities (one million) where civilian “clutter)
    • Only 1000 combat entities total
    • No combat entities used emiiters (no radar or laser)
the problem
The Problem
  • Simulation did not contain enough combat elements to effectively model theater wide combat.
    • More active entities are needed.
    • Modern combat vehicles make extensive use of EMF spectrum.
  • Radar and laser is a major component of a modern battlefield
    • Precision guided weapons
    • Active Radar Tracking/Radar Countermeasures
problem with emitters
Problem with Emitters
  • Under the current RTI-s interest declaration protocol there is only one emitter interest group
    • If you an interested in any emission you must receive all emissions
    • This is not a scalable solution.
proposed solution
Proposed Solution
  • Break up emitter interest group into geographic regions.
    • Simulations nodes will be allowed to specify specific geographic regions of interest and only receive emitter traffic from those regions
    • This forces receivers to compute the regions they are able to receive from before they can receive any traffic.
    • This implementation will be done the same as entity position interest groups are handled
      • Battle space is broken in a grid and each grid point is given a unique identifier
      • Receivers publish a list of grid square that they are interested in.
test bed
Test Bed
  • Computer simulation based on the 2005 large simulation run and personal experience with battlespace simulations
    • 100 nodes in Mesh network
    • Entities distributed evenly throughout nodes
      • This is done for real simulations to prevent computational “hotspots”
    • Only Emitter traffic monitored
    • 4 different distributions of emitter entities “locations” modeled
      • Broadcast – This is the current method. Entity location is irrelevant because controlling node will broadcast to all nodes
      • Random distribution – Entities in each node are randomly assigned a grid location
      • Host Cluster – Entities in the same node have a high probability of being in the same grid.
        • This is closest to how real simulations work
        • Two different distributions where tried. The probability of entities grouping was changed
          • 50% probability – half of the entities on the node where in the same grid
          • 30% probability – 30 % of the entities on the node where in the same grid
ad