By Vanessa Newey

1
Using Projection Aggregations to Support
Scalability in Distributed Simulation

• By Vanessa Newey

2
Introduction

• Background
• Scalability in Distributed Simulation
• Traditional Aggregation Techniques
• Problems with Traditional Methods
• Projection Aggregation
• Future Directions
• Summary

3
Distributed Interactive Simulation Systems

• Include
• Multiplayer Video Games,
• Collaborative Engineering
• Military and Industrial training
• Systems are Growing
• Can be hundreds of thousands of dynamic entities

4
(No Transcript)
5
Scalability in Distributed Simulation

• Number of entities that may simultaneously
  participate in the system.
• Entity participating object that is separately
  modeled.
• Scalability depends on
• Network Capacity
• Processor capabilities
• Rendering speeds
• Speed of throughput of shared servers

6
Limits to Scalability

• Host has to
• Receive updates
• Model and render scene
• Other tasks, including collision detection
• Soas the number of entities increases
• Increased Load on Network Resources.
• Increased Computational Load

7
Limits to Scalability cont.

• Even with dead-reckoning on a large simulation
  gt125,000 packets/second
• Approaching limit of interrupts for a
  general-purpose processor
• Rapidly increasing Computational Load due to
• Increasing number of entities
• Increasing detail of entity models
• More fine-grained graphical representations

8
Aggregation

• Used to reduce network and computational load.
• An aggregations is a simulation entity that
  represents a group of other entities.
• Previously used aggregation techniques are
  organisational and grid location

9
Organisational Aggregation

• Groups entities by their organisational
  structure. e.g. armies, brigades, battalions etc
• Easy to construct
• An organisations members may be dispersed
  throughout regions of the VR world
• In military simulations up to half the entities
  are destroyed
• Each host has to receive information from all
  organisation represented within that region
• Does not sufficiently reduce network traffic or
  the computational load

10
Grid Aggregation

• Groups entities by their location within the
  virtual world.
• Virtual world divided into regions, each is
  associated with an aggregation that transmits
  information about entities in that region
• Masks organisational structure
• For a host to send summary information about a
  battalion of tanks, then it must subscribe to
  information from all regions that could
  potentially contain one of those tanks.

11
Grid Aggregation cont.

• Does not allow remote hosts to receive different
  fidelity information for different entity types.
• Does not allow hosts to access entities by their
  organisation or type, so limited value for
  reducing network traffic or computational load.

12
Projection Aggregation

• Unifies the organisational and grid aggregation
  strategies.
• Each projection aggregation includes entities
  from a single organisation located within a
  single grid of the virtual world.
• Oraganisation is projected onto the virtual
  world.

13
(No Transcript)
14
Projection Aggregation cont.

• Each periodically transmits summary information
  about its members across the network.
• Hosts use projections to represent groups of
  remote entities that do not merit local modeling
  at high detail because far from viewer or
• Useful abstraction for describing all entities in
  a simulation

15
Projection Aggregation Summary Protocol

• Each periodically summarises its member entities
• Transmits summary packets over an associated
  multicast address.
• Enough information in packets for remote hosts to
  generate a low fidelity model of those member
  entities.
• Transmitted relatively infrequently (2-3
  seconds)
• Regular entities transmit every second
• Errors not noticeable

16
Projection Aggregation Summary Packets

• Contain
• count of entities represented
• a single position point (average position)
• Radius of bounding sphere
• Distribution information (mean, standard
  deviation)
• On remote host
• to place each entity the simulation generates a
  distribution of locations within the bounding
  sphere, given the mean and standard deviation.
• Projection aggregations center point and mean
  distance can be dead reckoned.

17
(No Transcript)
18
All entities treated as Projection Aggregation

• All entities can be treated as projection
  aggregations
• Only the rendering algorithm is different
• Remote hosts can dynamically change the fidelity
  of their local entity model without impacting
  other hosts
• Projections are easily integrated into existing
  simulations.
• Provides a natural mechanism for introducing more
  detailed entity models into an existing
  simulation.

19
(No Transcript)
20
(No Transcript)
21
Heirachical Projection Aggregation

• Projection Aggregations are arranged in a
  heirachy
• A projection aggregation is associated with a
  parent aggregation (a broader organisation and a
  larger grid.
• Each organisation maintains links to all its
  descendent entities.
• Grids maintain links to all projections in that
  grid
• Allows top-level filtering, reducing processing
• Collision detection algorithms can use projection
  aggregations to quickly filter unlikely or
  uninteresting collisions.

22
(No Transcript)
23
(No Transcript)
24
(No Transcript)
25
Experience with Projection Aggregations

• Used in PARADISE distributed simulation to assess
  the effectiveness of projections
• 72 less packets.
• Hosts multicast subscriptions down by 40
• Implemented and managed in roughly 4000 lines of
  code

26
Conclusion

• Projections provide a single abstraction for all
  simulation entities.
• Design allows hosts to view each entity in
  several ways
• Part of an organisation
• Part of a world Grid
• Part of a projection aggregation summary
• Projection aggregations reduce both network and
  computational requirements.
• A promising technique for network and
  computational resources
• support the evolution of more detailed entity
  models

27
More recently

• MASSIVE-3 uses abstraction as well, but more
  spatially focused
• In Singhals book he describes a slightly
  different approach. Where objects subscribe
  information sets.

28
(No Transcript)
29
(No Transcript)
30
(No Transcript)
31
(No Transcript)