Bill%20Helfinstine

1
Experiences with DataDistribution Managementin
Large-Scale Federations
Joint Experimentation and how it takes advantage
of DDM and IP Multicasting

• Bill Helfinstine
• Mark Torpey
• Deborah Wilbert
• Wayne Civinskas
• Lockheed Martin Information Systems

2
What is Joint Experimentation and what makes it
difficult?

• Exploration of possible changes in doctrine,
  organization, training, material, leadership and
  people to improve joint operations
• In order to explore jointness in an interesting
  way, a large scenario is needed
• Large geographic playbox
• Large number of vehicles to populate the playbox
• In order to support operators in the loop,
  real-time operation is needed
• In order to support large scenarios, large
  amounts of hardware and therefore space are
  needed
• Simulation distributed around the country to
  reduce cost

3
What do we mean by Large-scale federations

• Large number of objects (gt 30,000)
• Large number of federates (gt 100)
• Large number of sites around the country (5 or
  more)
• Simulation synchronized with real-time clock
• Therefore, there is a very large amount of data
  in playwhich must be handled quickly and
  efficiently

4
How can DDM help?

• DDM provides a means to tell the RTI what the
  federate is interested in.
• The RTI will do filtering on the data before the
  federate needs to get involved
• Provides a means to dynamically control the flow
  of data to the simulations
• The dynamicism of DDM provides a way to adapt to
  the changing data requirements of each simulation
  as the scenario progresses
• In a joint exercise, DM is often not powerful
  enough to describe the data requirements fully

5
Example
D
B
C
A

• Geographic interest regions

6
RTI implementations of DDM

• Distribute all subscription region information
  around to all federates, and do source filtering
• Include publication region information with each
  update/interaction and do receiver filtering
• Snap publication region to a grid and use the
  grid to do coarse filtering using the network
  itself
• With a gridded implementation of the DDM, IP
  multicasting can be employed
• IP multicast lets some or all of the filtering be
  done by the network hardware and operating
  system, to reduce load on the federates process.
• Just doing coarse filtering breaks the spec

7
How has DDM helped Joint Experimentation?

• In previous JE events, total object count was
  much higher than a single federate could accept.

MaxObjects PerFederate
Object Count
Vehicle Count
Federate Count
Event
400
7000
3600
300
STOW97
5000
40000
20000
100
J9901
20000
160000
100000
150
AO00
8
MC02 challenges for DDM use

• Millennium Challenge 2002 will be run using
  RTI-NG, which does perfect filtering on receive,
  which interacts poorly with our historical use of
  DDM
• We have not come up with a solution yet that
  gives us the flexibility and scalability we have
  enjoyed in the past, due to RTI-s and its tuning
  specifically for scalability and its
  implementation of inset grids
• JSAFs Hyperspace DDM scheme is hard to explain
  to other federate developers
• Some sort of DDM will probably be necessary due
  to the data requirements (gt30,000 objects at gt10
  sites)
• Gateways to DIS or private protocols make DDM
  work very poorly

9
Caveats of using DDM

• The design of the DDM scheme is fairly
  scenario-specific
• The operational design of the simulation usually
  does not take DDM into account
• Placement of opfor
• Centralized data collection
• Operational split by service means joint
  interaction happens over the WAN
• Many federates are designed to require all the
  data
• It is not easy to implement DDM support in a
  federate
• IP multicasting is not well supported by quite a
  bit of networking hardware

10
Caveats continued

• Loggers and wide-range viewers make DDM
  ineffective
• Since they need to receive almost everything, DDM
  provides no filtering
• Clever data collection design can reduce the data
  an individual logger receives
• Exercise controllers really like map displays
  with everything on them
• The HLA 1.3 spec has a number of requirements
  that we have heretofore ignored, which
  significantly reduce flexibility and performance