Overlay Topology Design Tradeoffs
Transcript
Overlay Topology Design Tradeoffs
QoE in Pull Based P2P-TV Systems: Overlay Topology Design Tradeoffs R. Fortuna, E. Leonardi, M. Mellia, M. Meo, S. Traverso IEEE International Conference on Peer-to-Peer Computing August 2010 S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 Outline Outline S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 1 / 29 Outline Outline 1 Introduction Motivations Definitions 2 Overlay Topology Design The Strategy Chunk Signaling and Scheduling Performance Evaluation Model Network Scenario Video Parameters Results 3 Video-Aware Schedulers The Strategy Results 4 Conclusions 5 Q&A S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 2 / 29 Introduction Introduction S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 3 / 29 Introduction Motivations Why this work? P2P-TV is a promising technology to reduce the cost of streaming content over the Internet, while offering a world-wide service... ... but many issues about how to improve the user perceived video quality are still open. The “Network-Aware P2P-TV Application over Wise Networks” FP7 project aims at developing an application to broadcast high definition video. We focused our attention on i) the design of the overlay topology, i.e. the network built by peers at application layer; ii) exploiting properties of encoded videostream to improve users’ QoE. S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 4 / 29 Introduction Motivations Why this work? P2P-TV is a promising technology to reduce the cost of streaming content over the Internet, while offering a world-wide service... ... but many issues about how to improve the user perceived video quality are still open. The “Network-Aware P2P-TV Application over Wise Networks” FP7 project aims at developing an application to broadcast high definition video. We focused our attention on i) the design of the overlay topology, i.e. the network built by peers at application layer; ii) exploiting properties of encoded videostream to improve users’ QoE. S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 4 / 29 Introduction Definitions Basic concepts A source peer splits the video streams in small chunks that are injected inside the overlay. Peers start exchanging chunks according to some scheduling scheme. Chunks are exchanged among peers that are neighbors of each others. P2P streaming systems, differently from file sharing, have to face with the real-time constraint! I Every chunk must be received within a deadline Dmax also called playout delay. S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 5 / 29 Introduction Definitions Assumptions Peers are Internet nodes and tipically their upload bandwidth is much lower than the downlink one. Peers upload bandwidth and latencies between peers are supposed to be known somehow. We neglect the effect of churning since scheduling dynamics are much faster. Peers exchange signaling information to trade chunks. S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 6 / 29 Introduction Definitions Assumptions Peers are Internet nodes and tipically their upload bandwidth is much lower than the downlink one. Peers upload bandwidth and latencies between peers are supposed to be known somehow. We neglect the effect of churning since scheduling dynamics are much faster. Peers exchange signaling information to trade chunks. S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 6 / 29 Introduction Definitions Assumptions Peers are Internet nodes and tipically their upload bandwidth is much lower than the downlink one. Peers upload bandwidth and latencies between peers are supposed to be known somehow. We neglect the effect of churning since scheduling dynamics are much faster. Peers exchange signaling information to trade chunks. S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 6 / 29 Introduction Definitions Assumptions Peers are Internet nodes and tipically their upload bandwidth is much lower than the downlink one. Peers upload bandwidth and latencies between peers are supposed to be known somehow. We neglect the effect of churning since scheduling dynamics are much faster. Peers exchange signaling information to trade chunks. S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 6 / 29 Overlay Topology Design Overlay Topology Design S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 7 / 29 Overlay Topology Design The Strategy Intuition To avoid long trading phases, peers with short end-to-end latencies should be connected to each other. To speed up chunk replication, high bandwidth peers should be well connected to each other (to the source). The number of neighbors of a peer have to choose, should be related to its upload bandwidth. S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 8 / 29 Overlay Topology Design The Strategy Intuition To avoid long trading phases, peers with short end-to-end latencies should be connected to each other. To speed up chunk replication, high bandwidth peers should be well connected to each other (to the source). The number of neighbors of a peer have to choose, should be related to its upload bandwidth. S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 8 / 29 Overlay Topology Design The Strategy Intuition To avoid long trading phases, peers with short end-to-end latencies should be connected to each other. To speed up chunk replication, high bandwidth peers should be well connected to each other (to the source). The number of neighbors of a peer have to choose, should be related to its upload bandwidth. S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 8 / 29 Overlay Topology Design The Strategy The Strategy How many neighbors? Being Kp the number of neighbors of a peer p chooses, we set Kp = max(3, 10 dBp /rs e) where Bp is the upload bandwidth of peer and rs is the average video rate of the stream. How to choose neighbors? Given peer p, all peers q such that Bq > 1/2Bp and lpq < 1/2E [lpq ] are marked as desired peers. Given the size of Kp , αKp peers are randomly selected and (1 − α)Kp peers are selected within a set of desired peers of p. The neighborood size The resulting size of the neighborood of a peer is Cp > Kp (edges are bidirectional). S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 9 / 29 Overlay Topology Design Chunk Signaling and Scheduling A pull mechanism Every peer periodically generates offer messages to publish the list of its useful chunks. Neighbors reply to each offer with a select message in which they specify the chunk they need. Once the select message is received, the chunk is then transmitted. When the chunk is received, an acknoledgement message is sent back to the trasmitter. P5 P1 P7 P2 OFFERS P4 P2 SELECTS Np Np OFFERS SELECTS time Chunk #1 to Peer 2 S. Traverso (Politecnico di Torino) Chunk #1 to Peer 5 Chunk #2 to Peer 1 Chunk #2 to Peer 2 Chunk #2 to Peer 7 Overlay Topology Design Tradeoffs Chunk #3 to Peer 4 August 2010 10 / 29 Overlay Topology Design Chunk Signaling and Scheduling Peer and Chunk Selections The Peer Selection Np neighbors are contacted by a peer in every offer session. Np is a fraction of the total neighborood Cp . In our tests, neighbors to contact with offer messages are chosen uniformly at random. The Chunk Selection Chunks requested in select messages are randomly chosen over the the list of useful ones proposed in offer messages. This is commonly known as Random Peer, Random Useful Chunk scheduling policy. S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 11 / 29 Overlay Topology Design Performance Evaluation Model Nework Scenario Our simulation involved 2000 peers partitioned in four classes according to their upload capacity: Class 1: 5.0Mb/s ± 10%; Class 2: 1.6Mb/s ± 10%; Class 3: 0.64Mb/s ± 10%; Class 4: 0Mb/s. We consider 4 scenarios, with increasing heterogeneity. The average upload bandwidth is E [Bp ] = 1.3Mb/s in all cases. Class H = 0.01 H = 0.05 H = 0.10 H = 0.15 S. Traverso (Politecnico di Torino) 1 1 5 10 15 2 76.7 58.5 35.8 13.2 3 2.3 16.5 34.2 51.8 Overlay Topology Design Tradeoffs 4 20 20 20 20 August 2010 12 / 29 Overlay Topology Design Performance Evaluation Model The Transport Network The transport network introduces a latency lpq to all the datagrams sent from peer p to q. Peers are distributed over the Earth surface and scattered over domains representing continents. Then, latencies are proportional to the geodetical distance between peers. 1 80 40 20 PDF Latitude [Deg] 60 0 -20 -40 -60 -80 -150 -100 -50 0 50 Longitude [Deg] 100 150 Figure: Peers distribution over Earth surface. S. Traverso (Politecnico di Torino) 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0 0.05 0.1 0.15 0.2 Latency [s] 0.25 0.3 0.35 Figure: Latency frequencies, E [lpq ] = 96ms. Overlay Topology Design Tradeoffs August 2010 13 / 29 Overlay Topology Design Performance Evaluation Model Real video sequences Three well-known video sequences have been considered as benchmarks. Pink of the Aerosmith Paris Foreman Length 40s 33.3s 40s Spatial Res. 352 × 240 352 × 288 352 × 288 Frame/sec 25 30 25 The videos consists of 1000 frames ≈ 40s of visualization. H.264/AVC codec has been adopted for encoding sequences. Hierarchical structure of GOP: frames can be IDR, P, B or b. Intra frames (IDR) carry valuable information (bigger), inter frames (P,B or b) carry differential information (and are much smaller). S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 14 / 29 Overlay Topology Design Results The Impact of α 45 PSNR [dB] 40 35 30 25 EVQ Dmax=6s Dmax=5s Dmax=4s Dmax=3s 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 α 1 Average PSNR for different values of α for ρ = 0.9 (ρ = E [Bp ]/rs ). S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 15 / 29 Overlay Topology Design Results 80 80 60 60 60 40 40 40 20 0 -20 -40 20 0 -20 -40 -60 -60 -80 -80 -150 -100 -50 0 50 Longitude [Deg] 100 150 Latitude [Deg] 80 Latitude [Deg] Latitude [Deg] The Impact of α II 20 0 -20 -40 -60 -80 -150 -100 -50 0 50 Longitude [Deg] 100 Figure: α = 0.0 Figure: α = 0.1 S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs 150 -150 -100 -50 0 50 Longitude [Deg] 100 150 Figure: α = 1.0 August 2010 16 / 29 Overlay Topology Design Results 80 60 60 40 40 40 20 0 -20 -40 20 0 -20 -40 -60 -60 -80 -80 -150 -100 -50 0 50 Longitude [Deg] 100 Latitude [Deg] 80 60 Latitude [Deg] 80 150 Figure: α = 0.0 20 0 -20 -40 -60 -80 -150 -100 -50 0 50 Longitude [Deg] 100 150 Figure: α = 0.1 -150 -100 -50 0 50 Longitude [Deg] 100 150 Figure: α = 1.0 46 44 PSNR [dB] Latitude [Deg] The Impact of α II 42 40 EVQ α=0.0 Dmax=5s α=0.1 Dmax=5s α=1.0 Dmax=5s 38 36 1 6 11 16 21 26 Peers ID 31 36 41 α = 0.0 can lead to disconnected topologies! S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 16 / 29 Overlay Topology Design Results 45 45 40 40 PSNR [dB] PSNR [dB] Adapting Kp to the upload capacity 35 30 25 EVQ Dmax=6s Dmax=5s 35 30 Dmax=4s Dmax=3s 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 α 1 25 EVQ Dmax=6s Dmax=5s Dmax=4s Dmax=3s 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 α 1 Fixed Kp = 20 (left) and variable Kp = max(3, 10 dBp /rs e) (right) for ρ = 0.9. S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 17 / 29 Overlay Topology Design Results Network stress 0.1 Network stress [ms] 0.09 0.08 0.07 0.06 0.05 Dmax=3s Dmax=4s Dmax=5s Dmax=6s 0.04 0.03 0.02 0 0.2 0.4 0.6 0.8 1 α Figure: The network stress index, i.e. the average distance covered by chunks expressed it terms of the corresponding latency (ρ = 0.9). S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 18 / 29 Overlay Topology Design Results Robustness 46 H=0.01 H=0.05 H=0.10 H=0.15 EVQ 44.5 44 43.5 43 42.5 42 41.5 41 0.025 45 PSNR [dB] PSNR [dB] 45.5 45 40 35 30 25 0.22 0.415 0.61 α 0.805 1 Paris Foreman Pink Paris EVQ Foreman EVQ Pink EVQ 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 α 1 Figure: Average PSNR versus α for Figure: Average PSNR versus α for different degrees of heterogeneity H with different video sequences with ρ = 0.9, H = 0.10 and variable Kp . Dmax = 5s and ρ = 0.9. S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 19 / 29 Video-Aware Schedulers Video-Aware Schedulers S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 20 / 29 Video-Aware Schedulers PSNR [dB] PSNR [dB] Different frame types and losses 55 50 45 40 35 30 25 20 15 55 50 45 40 35 30 25 20 15 0 5 10 15 20 25 Time [s] 30 35 40 0 5 10 15 20 25 Time [s] 30 35 40 Figure: PSNR variation of a random peer versus time. ρ = 0.6 (rs = 780kb/s, top) and ρ = 1.0 (rs = 1290kb/s, bottom). S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 21 / 29 Video-Aware Schedulers The Strategy Improving chunk selection Different frame types have different importance in a video stream. A frame loss may cause very different levels of degradation of the reconstructed video quality: I I missing an IDR frame impairs the video decoding until the next IDR; missing a b frame impairs only the decoding of one single frame. We can assign priority to chunks transporting precious frames. To avoid chopping frames into several chunks, we set one frame per chunk. Priority is assigned to frames based on the amount of degradation they might induce if lost: q is the number of subsequent frames that would be affected by the lost of that frame. A weight q ω is given to every chunk encapsulating a frame. With ω > 0 we assign a larger weight to more important frames. S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 22 / 29 Video-Aware Schedulers The Strategy Improving chunk selection Different frame types have different importance in a video stream. A frame loss may cause very different levels of degradation of the reconstructed video quality: I I missing an IDR frame impairs the video decoding until the next IDR; missing a b frame impairs only the decoding of one single frame. We can assign priority to chunks transporting precious frames. To avoid chopping frames into several chunks, we set one frame per chunk. Priority is assigned to frames based on the amount of degradation they might induce if lost: q is the number of subsequent frames that would be affected by the lost of that frame. A weight q ω is given to every chunk encapsulating a frame. With ω > 0 we assign a larger weight to more important frames. S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 22 / 29 Video-Aware Schedulers Results Slight Improvements 42 Pink PSNR [dB] 40 38 36 34 ω=0.0 ω=0.5 ω=1.0 ω=2.0 32 5 10 15 20 25 PeerID 30 35 40 45 Figure: PSNR for different peers and values of ω with ρ = 1.1, Dmax = 5s and H = 0.10. Pink by Aerosmith video sequence. S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 23 / 29 Video-Aware Schedulers Results Slight Improvements II 40 38 Paris PSNR [dB] 36 34 32 30 ω=0.0 ω=0.5 ω=1.0 ω=2.0 28 26 5 10 15 20 25 PeerID 30 35 40 45 Figure: PSNR for different peers and values of ω with ρ = 1.1, Dmax = 5s and H = 0.10. Paris video sequence. S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 24 / 29 Video-Aware Schedulers Results Slight Improvements III 38 PSNR [dB] 36 Foreman 34 32 30 ω=0.0 ω=0.5 ω=1.0 ω=2.0 28 26 5 10 15 20 25 PeerID 30 35 40 45 Figure: PSNR for different peers and values of ω with ρ = 1.1, Dmax = 5s and H = 0.10. Foreman video sequence. S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 25 / 29 Conclusions Conclusions S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 26 / 29 Conclusions Conclusions We provided guidelines for the design of the overlay topology and the chunk scheduling algorithm. By carefully designing the overlay topology we can partially localize the traffic and improve the user QoE. By prioritizing chunks that encapsulate valuable pieces of information at the scheduler level, system performance can be slightly improved in overloaded conditions. S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 27 / 29 Conclusions Conclusions We provided guidelines for the design of the overlay topology and the chunk scheduling algorithm. By carefully designing the overlay topology we can partially localize the traffic and improve the user QoE. By prioritizing chunks that encapsulate valuable pieces of information at the scheduler level, system performance can be slightly improved in overloaded conditions. S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 27 / 29 Conclusions Conclusions We provided guidelines for the design of the overlay topology and the chunk scheduling algorithm. By carefully designing the overlay topology we can partially localize the traffic and improve the user QoE. By prioritizing chunks that encapsulate valuable pieces of information at the scheduler level, system performance can be slightly improved in overloaded conditions. S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 27 / 29 Q&A Q&A S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 28 / 29 Q&A Thank you for your attention! S. Traverso (Politecnico di Torino) Overlay Topology Design Tradeoffs August 2010 29 / 29