Selection Strategies for PeertoPeer 3D Streaming

1
Selection Strategies for Peer-to-Peer 3D Streaming

• Wei-Lun Sung, Shun-Yun Hu, Jehn-Ruey Jiang
• National Central University, Taiwan
• 2008/05/29

2
Virtual environments (VE)

• VEs allow users to interact in synthetic worlds
• Larger content more worlds ? content streaming
  (i.e., 3D streaming) becomes necessary

3
3D streaming

• Continuous and real-time delivery of 3D content
  to allow user interactions without a full
  download.
• Object streaming fragments mesh into base
  refinements

Refinements
Base
1
2
3
(Hoppe 96)
User
4
Scene streaming

• multiple objects
• object selection prioritization

Teler Lischinski 2001
5
Comparison with media streaming

• Highly interactive (latency-sensitive)
• Behavior-based (non-linear)
• How to scale to millions of concurrent users?

6
Imagine you start with a globe
7
Zoom in
8
To a city
9
and a building
10
Right now its flat
11
But in the near future
12
Observation

• Limited predictable area of interest (AOI)?
• Overlapped visibility shared content

13
Benefits of peer-to-peer

• Scalable
• Growing amount of total resources
• Affordable
• Commodity PCs
• Feasible
• Better client hardware (CPU, broadband networks)?
• Availability of user-hosted machines

14
Peer selection

• Choose suitable candidates so that content
  retrieval can be done quickly and efficiently
• Source discovery
• Which peers possess the needed data
• Source selection
• Which peers to request the data

15
Related Work FLoD Infocom 2008

• VE partitioned into cells with scene descriptions
• Assumes P2P overlay that provides AOI neighbors

star self triangles neighbors circle
AOI rectangles objects
16
Peer selection in FLoD

• Source discovery
• Query-response
• Extra delay due to queries
• Source selection
• Random selection
• Requests contention due to overlapping requests

17
Request contention problem

• Overlapping requests create contentions

R6
R1
R5
R2
R4
R1,R2,R3,R4,R5,R6
OBJ
R3
R1,R2
R1,R2,R3
18

• Proposed Solutions

19
Incremental Piece List Exchange

• Proactive notification of content availability
• Periodic incremental exchange of content
  availability information with neighbors.

incremental content information
Msg_Type
Obj_ID
Max_PID
Obj_ID
Max_PID
????
20
Extended Candidate Buffer

• Non-AOI neighbors may still possess data
• Maintain extra list of non-AOI neighbors

S
R
Obj
21
Multi-Level AOI Request

• Localized requests may prevent contentions
• Peers request from closer neighbors/levels first

22
Simulation Environment

• Based on FLoD (available on SourceForge)
• World size 1000 x 1000
• Simulation steps 3000
• Objects 500
• Nodes 50 500 (50 nodes increase)
• AOI radius 75
• Server bandwidth 10 Mbps / 10 Mbps
• Peer bandwidth 1 Mbps / 256 Kbps

23
Simulation Environments (cont.)

• Source discovery
• (QR) query-response 5 steps interval, 10
  requests
• (EE) exchanged extended 150 radius
• Source selection
• (RAND) random
• (ML) multi-level AOI request 4 levels
• Original FLoD QR-RAND
• Proposed method EE-ML

24
Hit Ratio
25
Base Latency
26
Fill ratio
27
Bandwidth (Server)
28
Bandwidth (Clients source discovery)
29
Conclusion

• New selection strategies for P2P 3D streaming
• Availability info exchange extended candidate
  buffer reduce both latency and bandwidth overhead
• multi-level AOI requests obtain data from closer
  providers but improve only hit ratio
• Future work
• More sources
• Physical topology
• Pre-fetching

30

• Q A

31
Neighbor discovery via VON
Voronoi diagrams identify boundary neighbors for
neighbor discovery
Non-overlapped neighbors
Boundary neighbors
New neighbors
Hu et al. 06
32
LODDT
Object
Tree Node
Aura
33
LODDT
Object
Tree Node
Aura
U
34
LODDT (cont.)
Requests
Candidates

• Discovery
• Estimation
• Selection
• Every peer samples the time-to-serve (TTS) of its
  neighbors
• Requestors organize their data requests so as
  obtain tree nodes in the right order
• Drawback incorrect estimation, congestion

35
Simulation Environments (cont.)

• System performance
• Hit ratio Ratio of successful requests peers
  have sent
• Latency Duration between initial request and
  data arrival
• Fill ratio Ratio of the possessed required data
• Scalability metrics
• Bandwidth usage (consumption)
• Content discovery overhead