Upload
christian-timmerer
View
1.120
Download
0
Embed Size (px)
DESCRIPTION
Citation preview
AN EVALUATION OF PIECE-PICKING ALGORITHMS FOR LAYERED
CONTENT IN BITTORRENT-BASED PEER-TO-PEER SYSTEMS
ICME 2011
Special Session on Hot Topics in Multimedia Delivery
Michael Eberhard 1Piece-Picking Algorithm Evaluation
Michael Eberhard
Hermann Hellwagner
Christian Timmerer
AAU Klagenfurt
Tibor Szkaliczki Laszlo Szobonya
MTA SZTAKI
OVERVIEW
Introduction to Piece Picking
Algorithm for Layered Piece Picking
Evaluation Results
Single/Multi Layer Comparison
Michael Eberhard Piece-Picking Algorithm Evaluation 2
PIECE-PICKING IN P2P NETWORKS
When streaming layered videos in a P2P network, the piece-picking algorithm needs to decide which piece to download at which point in time.
The main goal is to provide the best possible quality with the available bandwidth while ensuring continuous playback and minimizing changes in quality.
Michael Eberhard Piece-Picking Algorithm Evaluation 3
PIECE-PICKING BUFFER
Michael Eberhard Piece-Picking Algorithm Evaluation 4
The piece-picking algorithm provides a download strategy for all pieces within the sliding window.
The sliding window contains the pieces that are required for playback in the near future.
PIECE-PICKING SLIDING WINDOW
Michael Eberhard Piece-Picking Algorithm Evaluation 5
PIECE UTILITY CALCULATION
For each piece within the sliding window, the utility is defined as
Michael Eberhard Piece-Picking Algorithm Evaluation 6
𝑼𝒊 𝒋 𝒌 = 𝐝𝐣 × 𝐝𝐩𝐢 𝐣 𝐤(𝐭𝐢 − 𝐭𝐤)𝛂
lj: the layer of the pieceti: the point in time at which the piece is displayedtk: the point in time of the actual decisiondj: the distortion reduction importance of the piecedpijk: the probability to receive the useful piece in time
PIECE MAPPING (1)
GOPs of 64 frames are considered as a unit
2.56 seconds of content are provided commonly for each layer (at 25 fps)
A unit is always entirely downloaded
Only supports layered scalability
For single layer content, ~16 frames of content are mapped to a unit
Michael Eberhard Piece-Picking Algorithm Evaluation 7
PIECE MAPPING (2)
Michael Eberhard Piece-Picking Algorithm Evaluation 8
SIMULATION SETUP
Omnet++/Oversim with new P2P protocol and applications (piece picking algorithms)
Swarm of 100 peers, streaming a one hour video
Peer arrivals and departures are modeled according to a poisson distribution
Michael Eberhard Piece-Picking Algorithm Evaluation 9
MULTI/SINGLE LAYER COMPARISON
Both are encoded with the same constant bitrate and split to fixed-size pieces
Quality for single layer is higher due to SVC overhead
Comparison based on PSNR, as piece size is equal for both encodings (received bitrate is ~equal)
The single layer PSNR for missing pieces is weighted with the PSNR of a black frames
Michael Eberhard Piece-Picking Algorithm Evaluation 10
FULL BANDWIDTH, NO CHURN
Michael Eberhard Piece-Picking Algorithm Evaluation 11
FULL BANDWIDTH, 10% CHURN
Michael Eberhard Piece-Picking Algorithm Evaluation 12
LIMITED BANDWIDTH, 10% CHURN
Michael Eberhard Piece-Picking Algorithm Evaluation 13
MIN/MAX QUALITY/PEER FOR SL
Michael Eberhard Piece-Picking Algorithm Evaluation 14
# Pieces %
4 82.16
3 16.63
2 1.17
1 0.04
0 0
FULL BANDWIDTH, INCREASING CHURN
Michael Eberhard Piece-Picking Algorithm Evaluation 15
FULL BANDWIDTH, 10% CHURN, FRAME LOSS
Michael Eberhard Piece-Picking Algorithm Evaluation 16
CONCLUSION
Layered Video Codecs can be integrated in Bittorrent-based P2P system in a backwards-compatible way
If the bandwidth conditions are not optimal, layered codecs provide a clearly better performance in terms of PSNR
Michael Eberhard Piece-Picking Algorithm Evaluation 17
THANK YOU FOR YOUR ATTENTION!
Michael Eberhard Piece-Picking Algorithm Evaluation 18