Research:%20Group%20communication%20in%20distributed%20interactive%20applications

1

• Research Group communication in distributed
  interactive applications
• Student Knut-Helge Vik
• Institute University of Oslo, Simula Research
  Labs

2
Outline

• MiSMoSS
• Motivation
• Group Communication
• Application Layer Multicast
• Tree algorithms
• Research
• Conclusions

3
MiSMoSS Project

• Investigate Large-scale interactive applications
• Main issue Latency
• Three sub-projects
• Latency hiding
• prediction
• Group communication management
• Overlay multicast
• Transmission protocol optimization
• Thin streams

4
Motivation - Group communication management

• Large-scale interactive applications
• Users interact in groups
• Communication demands vary within an application
• low latency demands
• high bandwidth demands
• frequent group membership changes
• consistency
• Build overlay routing
• with a small diameter
• with degree limitations
• using algorithms with low execution times
• with stable reconfigurations

5
Motivation

• Example application Massively Multiplayer Online
  Games
• Large scale (thousands of simultaneous users)
• Central server-based (experience high latency)
• Clients far apart in physical world, but near in
  virtual world
• Issues Event distribution
• Goal Reduce latency, decrease server load,
  increase MMOG size
• Group Communication Application Layer Multicast
• Overlay Multicast Must handle group dynamics
• Current overlay multicast protocols lack
  efficient dynamic handling
• Goal Create a dynamic overlay multicast protocol

Real-World Proximity
Virtual World Proximity
6
Research - Summary

• Group membership - join/leave
• Insert or remove group members to an existing
  topology
• Overlay multicast fully meshed graph
• Optimization techniques edge pruning, core
  selection
• Multicast trees
• Investigating tree problems
• Shortest path tree
• Minimum spanning tree
• Steiner minimum tree SPH, DNH, ADH
• Minimum diameter degree limited spanning tree
• Dynamic tree algorithms insert and remove
• Tree algorithm constraints
• unconstrained
• degree and/or delay constrained
• Metrics
• Stress - degree
• Diameter maximum pairwise latency
• Total tree cost sum of edge weights
• Reconfiguration time time it takes to complete
  reconfiguration

7
Research - Optimization

• Application layer graphs are fully meshed
• Ex V1000, E 499500 edges, E_T V - 1
  (using 0.02 of the edges)
• Tree algorithms build trees using graphs
• Graph optimization techniques
• Edge pruning algorithms k-Best links
• Limit nodes to group members steiner minimum
  trees?
• Core selection heuristics
• Include stronger nodes in the input graph
  higher stress capacity
• Especially suitable for SMT heuristics
• Group center, topological center, MDDL center
• Goal Reduce reconfigure time while preserving
  tree quality

8
Research - Group Dynamics

• Dynamic membership nodes join and leave the
  multicast tree dynamically
• Must insert and remove nodes online
• Needs algorithm to reconfigure the tree
• Contradictory goals
• Low reconfigure time
• efficient tree
• tree stability

9
Research Reconfiguration Set

• Reconfiguration set nodes involved in
  reconfiguration
• Entire group
• Pros Tree efficiency
• Cons High reconfiguration time, tree stability
• Reduced size of reconfiguration set
• Pros Low reconfiguration time, increased
  stability
• Cons Reduced tree efficiency

10
Research Reconfiguration Set

• Reconfiguration set nodes involved in
  reconfiguration
• Entire group
• Pros Tree efficiency
• Cons High reconfiguration time, tree stability
• Reduced size of reconfiguration set
• Pros Low reconfiguration time, increased
  stability
• Cons Reduced tree efficiency

11
Tree Algorithms

• Tree algorithms reconfigures entire tree
• Problems in P Minimum spanning tree (MST),
  Shortest path tree (SPT)
• Problems in NP Steiner minimum tree (SMT),
  Minimum diameter degree limited tree, Degree
  constrained MST, SPT, SMT
• Main issues reconfiguration time is high and
  tree stability suffers
• Heuristics are especially slow
• Addressing issues Reduce number of edges in
  input graph, include strong cores
• Pros Reduced reconfiguration time, increased
  stability
• Cons Tree efficiency is also reduced

Reconfiguration time
Total tree cost
12
Dynamic Algorithms

• Dynamic Algorithms insert/remove (reconfigure
  smaller parts of a tree)
• basic edge optimization goals Minimum cost edge,
  Minimum diameter edge, Minimum cost to source
• Prune non member nodes
• Main issues tree efficiency suffers
• Always local optimizations
• Crowded with non member nodes
• Addressing issues Vary reconfiguration set size,
  prune non-members, switch non members to stronger
  cores
• Cons Increased reconfiguration time, reduced
  stability
• Pros Tree efficiency

Edge changes remove algorithms (100 nodes)
13
Insert Algorithms

• Basic insertion choices
• Insert as leaf no edge change
• Insert and reconfigure increased tree
  efficiency but reconfiguration time!
• Implemented a number of insert algorithms ex
• I-MC insert minimum cost edge
• I-MDDL insert minimum diameter degree limited
  edge

Node is joining
Connect to tree as leaf
Insert strong core
Use as intersection
Three configuration examples
14
Remove Algorithm

• Basic remove choices
• Remove leaf no edge changes (easy)
• Remove non-leaf MUST reconfigure
• reconfigure and add/remove non-MN
• Implemented a number of algorithms ex
• RTR-MC neighbors
• RTR-P pruning non members

Keep as non-member
Use stronger core
Reconnect neighbors
Node is leaving
Three configuration examples
15
Dynamic Algorithms Insert/Remove
reconfigure set
leaving
reconfigure set
RTR-MC
RTR-P
16
Conclusions and Future Work

• Current algorithms are centralized
• Implement distributed algorithms
• PlanetLab implementation
• Implement overlay multicast protocol
• Investigate mesh vs. trees
• Questions?