Download presentation
Presentation is loading. Please wait.
1
M3: Practical and Reliable Multi-Layer Video Multicast over Multi-Rate Wi-Fi Network
Menghan Li*, Dan Pei, Xiaoping Zhang Beichuan Zhang, Zhi Wang, Hailiang Xu, Zihan Wang Thank you for the introduction. Good morning, everyone! My name is limenghan and I’m from Tsinghua University. Today I would like to present my paper “…” First, I’m going to introduce the research background of this work. 2018/12/3 IWQoS 2016
2
Video Multicast over Wi-Fi
Wi-Fi is a universal solution for the last-hop access networks Account for 66% of IP traffic by 2020 Global public Wi-Fi hotspots: 7.8 million in 2015 In large and public assembly places Multiple users view the same popular events via the same APs Unicast-based solution does not scale and wastes bandwidth Wi-Fi has been a universal solution for the last-hop access networks. With the rapid growth of mobile devices and the wide deployment of Wi-Fi hotspots, in large and public assembly places, people may view the same popular events via the same APs. In this scenario, unicast-based solution does not scale and wastes a lot of bandwidth 2018/12/3 IWQoS 2016
3
Video Multicast over Wi-Fi
Wi-Fi is a universal solution for the last-hop access networks Account for 66% of IP traffic by 2020 Global public Wi-Fi hotspots: 7.8 million in 2015 In large and public assembly places Multiple users view the same popular video via the same APs Unicast-based solution does not scale and wastes bandwidth Supporting video multicast over deployed Wi-Fi networks Therefore, supporting video multicast over deployed Wi-Fi networks faces the actual demand. 2018/12/3 IWQoS 2016
4
Wi-Fi’s multi-rate nature
90Mbps C2 AP/802.11n 150Mbps C1 6.5Mbps Due to Wi-Fi’s multi-rate nature, clients with different channel qualities can support different PHY rates. C3 -50dBm -55dBm -80dBm 2018/12/3 IWQoS 2016
5
Limitations of single-layer video multicast
The video quality is limited by the client with the worst channel quality 90Mbps C2 AP/802.11n 150Mbps C1 2000Kbps 4000Kbps 8000Kbps H.264/AVC 6.5Mbps Therefore, in traditional single-layer-coded video streams, all clients have to settle with the lowest video rate limited by the client with the worst channel quality. C3 -50dBm -55dBm -80dBm 2018/12/3 IWQoS 2016
6
Scalable Video Coding (SVC)
SVC is a multi-layer-coded video coding scheme One base layer with the lowest video quality Several enhancement layers that can increase video frame rate, resolution or picture quality BL EL1 EL2 SVC SVC is a multi-layer-coded video coding scheme SVC video stream has a mandatory base layer with the lowest video quality, and several optional enhancement layers that can increase video frame rate, resolution or picture quality. 2018/12/3 IWQoS 2016
7
SVC video multicast over Wi-Fi
BL EL1 EL2 SVC 90Mbps EL1 C2 EL2 AP/802.11n BL It is natural to use SVC and enable the AP to multicast the mandatory base layer at the lowest bit rate supported by the client with the worst channel, and multicast optional enhancement layers at higher bitrates supported by those clients with better channel qualities. There have been several papers to propose SVC video multicast with rate adaptation mechanisms. C1 6.5Mbps C3 -50dBm -55dBm Lim et al., TMC 2012 Kuo et al., TCSVT 2015 -80dBm 2018/12/3 IWQoS 2016
8
SVC video multicast over Wi-Fi
BL EL1 EL2 SVC 90Mbps EL1 Need to modify MAC protocol Impractical to be deployed C2 EL2 AP/802.11n BL However, these solutions need to modify MAC protocol and are impractical to be deployed. C1 6.5Mbps C3 -50dBm -55dBm Lim et al., TMC 2012 Kuo et al., TCSVT 2015 -80dBm 2018/12/3 IWQoS 2016
9
Pseudo-broadcast (PB)
A multicast approach based on Wi-Fi unicast No need to modify MAC protocol The AP explicitly selects a client as the unicast receiver Others clients listen for the packets in the promiscuous mode AP Receiver Listener Unicast Sniff Pseudo-broadcast is a multicast approach based on wifi unicast, it does not need to modify mac protocol. The basic ideas are, The AP explicitly select a client as the unicast receiver, usually, this receiver is the client with the worst channel quality. And other clients listen for the packets in the promiscuous mode. 2018/12/3 IWQoS 2016
10
Pseudo-broadcast (PB)
A multicast approach based on Wi-Fi unicast No need to modify MAC protocol The AP explicitly selects a client as the unicast receiver Others clients listen for the packets in the promiscuous mode Only PB-based single-layer video multicast Low overall video quality AP Receiver Listener Unicast Sniff However, existing PB-based solutions selecting one receiver only support single-layer video multicast and still suffer from low overall video quality. Chandra et al., ICNP 2009 Sen et al., NSDI 2010 2018/12/3 IWQoS 2016
11
Design Goal Limitations of existing solutions
Being impractical to deploy (Modify protocol) Low overall video quality (Single-layer video multicast) In summary, existing approaches suffer from either being impractical to deploy and low overall video quality in existing multi-rate Wi-Fi networks. 2018/12/3 IWQoS 2016
12
Design Goal Limitations of existing solutions
Being impractical to deploy (Modify protocol) Low overall video quality (Single-layer video multicast) M3: Multi-layer video Multicast with Multi-receiver pseudo- broadcast Practical: No change to protocol Reliable: All clients can view the video smoothly Optimized: Maximize the overall video quality Therefore, we propose M3 and use multi-receiver pseudo-broadcast to realize a practical and reliable multi-layer wifi video multicast solution. Our design goal is to maximize the overall video quality received by all clients. 2018/12/3 IWQoS 2016
13
Outline Background Challenges M3 system design Results Conclusion
In the following, I will present some design challenges faced by M3 2018/12/3 IWQoS 2016
14
Challenge 1: How to improve pseudo-broadcast’s reliability
PB cannot guarantee the reliability for listeners The first challenge is how to improve … We use the following results to verify that PB cannot guarantee the reliability for listeners. The mean value of minimum packet delivery ratio in one block Block size is 100 packets 2018/12/3 IWQoS 2016
15
Challenge 1: How to improve pseudo-broadcast’s reliability
PB cannot guarantee the reliability for listeners Receiver’s RSSI is more lower than listener’s RSSI Higher probability to receive the packets From this figure, we can see that when the receiver’s RSSI is more lower than the listener’s RSSI, the listener has a higher probability to receive packets. The mean value of minimum packet delivery ratio in one block Block size is 100 packets 2018/12/3 IWQoS 2016
16
Challenge 1: How to improve pseudo-broadcast’s reliability
PB cannot guarantee the reliability for listeners RSSI -50 -55 -60 -65 -70 -75 -80 -85 50.1 51.6 45.8 30.2 3.5 52.3 50.5 49.2 14.3 1.6 55.9 60.2 46.4 28.4 3.8 0.9 0.2 -66 72.8 75.5 62.4 47.4 20.9 6.6 92.1 94.1 88.9 80.9 41.0 23.8 90.1 96.5 87.9 84.1 83.5 53.4 15.7 95.1 96.3 89.6 91.4 94.3 89.0 45.9 12.9 95.7 99.0 93.7 96.7 96.9 90.7 56.3 And from this table, we can get more detailed results. Even if the receiver has the lowest RSSI, it is difficult to ensure that the listener has a PDR of 100%. Difficult to ensure a PDR of 100% 2018/12/3 IWQoS 2016
17
Challenge 1: How to improve pseudo-broadcast’s reliability
PB cannot guarantee the reliability for listeners RSSI -50 -55 -60 -65 -70 -75 -80 -85 50.1 51.6 45.8 30.2 3.5 52.3 50.5 49.2 14.3 1.6 55.9 60.2 46.4 28.4 3.8 0.9 0.2 -66 72.8 75.5 62.4 47.4 20.9 6.6 92.1 94.1 88.9 80.9 41.0 23.8 90.1 96.5 87.9 84.1 83.5 53.4 15.7 95.1 96.3 89.6 91.4 94.3 89.0 45.9 12.9 95.7 99.0 93.7 96.7 96.9 90.7 56.3 Choose the receiver whose RSSI is lower than the listener’s RSSI To improve PB’s reliability, first, the sender must choose the receiver … Difficult to ensure a PDR of 100% 2018/12/3 IWQoS 2016
18
Challenge 1: How to improve pseudo-broadcast’s reliability
PB cannot guarantee the reliability for listeners RSSI -50 -55 -60 -65 -70 -75 -80 -85 50.1 51.6 45.8 30.2 3.5 52.3 50.5 49.2 14.3 1.6 55.9 60.2 46.4 28.4 3.8 0.9 0.2 -66 72.8 75.5 62.4 47.4 20.9 6.6 92.1 94.1 88.9 80.9 41.0 23.8 90.1 96.5 87.9 84.1 83.5 53.4 15.7 95.1 96.3 89.6 91.4 94.3 89.0 45.9 12.9 95.7 99.0 93.7 96.7 96.9 90.7 56.3 A combined application-layer Forward Error Correction (FEC) and Automatic Repeat-reQuest (ARQ) mechanism Then, we use a combined application-layer FEC and ARQ mechanism to further reduce the data loss for listeners. Difficult to ensure a PDR of 100% 2018/12/3 IWQoS 2016
19
Challenge 2: How to choose receivers and assign SVC layers
Maximize the overall video quality with limited bandwidth Video rate of each SVC layer Goodput of each client 𝐺𝑃 2 C2 AP 𝐺𝑃 1 BL EL1 EL2 SVC C1 The second challenge is how to choose … When we get the video rate of each SVC layer and the goodput of each client, we need to assign SVC layers to proper receivers, and our aim is to maximize the overall video quality with limited bandwidth. 𝑉𝑅 1 𝐺𝑃 3 C3 -50dBm 𝑉𝑅 2 -55dBm 𝑉𝑅 3 -80dBm 2018/12/3 IWQoS 2016
20
Challenge 2: How to choose receivers and assign SVC layers
Use binary integer linear programming to derive an optimal map between SVC layers and receivers 1 2 L SVC layers Receivers R1 R2 RH Here, we use binary integer linear programming to derive an optimal map between SVC layers and receivers. The target is to maximize the total video rate received by all clients. Maximize the total video rate received by all clients 2018/12/3 IWQoS 2016
21
Challenge 3: How to adapt to the network dynamics
Measure the goodput of each receiver 𝐺𝑃=𝑇𝑃∙ 𝑃𝐷𝑅 𝑚𝑖𝑛 𝑃𝐷𝑅 4 𝑃𝐷𝑅 2 C4 C2 Listener Listener 𝑇𝑃 1 C1 𝑃𝐷𝑅 3 AP The third challenge is how to adapt … To measure the receiver’s goodput, we need to get the receiver’s TCP throughput and all other listener’s pdr. C3 Receiver Listener 2018/12/3 IWQoS 2016
22
Challenge 3: How to adapt to the network dynamics
Measure the goodput of each receiver 𝐺𝑃=𝑇𝑃∙ 𝑃𝐷𝑅 𝑚𝑖𝑛 𝑃𝐷𝑅 4 / 𝑃𝐷𝑅 4 ′ 𝑇𝑃 2 / 𝑃𝐷𝑅 2 C4 C2 Listener Receiver/Listener 𝑇𝑃 1 C1 𝑃𝐷𝑅 3 / 𝑃𝐷𝑅 3 ′ AP To optimize the overall video quality, we need to get the goodput of each potential receiver. C3 Receiver Listener 2018/12/3 IWQoS 2016
23
Challenge 3: How to adapt to the network dynamics
Measure the goodput of each receiver 𝐺𝑃=𝑇𝑃∙ 𝑃𝐷𝑅 𝑚𝑖𝑛 𝑃𝐷𝑅 4 / 𝑃𝐷𝑅 4 ′ 𝑇𝑃 2 / 𝑃𝐷𝑅 2 C4 C2 C5 Listener Receiver/Listener 𝑇𝑃 1 C1 𝑃𝐷𝑅 3 / 𝑃𝐷𝑅 3 ′ AP And when the new client arrives or other network conditions change, we only have a little time to re-evaluate system settings to approach the optimized target. C3 Receiver Listener 2018/12/3 IWQoS 2016
24
Challenge 3: How to adapt to the network dynamics
Measure the goodput of each receiver 𝐺𝑃=𝑇𝑃∙ 𝑃𝐷𝑅 𝑚𝑖𝑛 𝑃𝐷𝑅 4 / 𝑃𝐷𝑅 4 ′ 𝑇𝑃 2 / 𝑃𝐷𝑅 2 A dynamic feedback mechanism to collect some necessary statistics from clients C4 C2 C5 Listener Receiver/Listener 𝑇𝑃 1 C1 𝑃𝐷𝑅 3 / 𝑃𝐷𝑅 3 ′ AP Here, we use a dynamic feedback mechanism to collect some necessary statistics from clients. C3 Receiver Listener 2018/12/3 IWQoS 2016
25
Outline Background Challenges M3 system design Results Conclusion
In the following, I will present M3 system design in detail. 2018/12/3 IWQoS 2016
26
M3 system design C2 C1 Challenges Solutions AP …… Pseudo-broadcast
Practicability AP Reliability First, to be practical to deploy in existing wifi networks, M3 uses pseudo-broadcast to perform video multicast. SVC layer 1 SVC layer 2 SVC layer 3 …… Optimization Dynamics 2018/12/3 IWQoS 2016
27
M3 system design C2 C1 Challenges Solutions AP …… Pseudo-broadcast
layer1 FEC layer2 Practicability FEC AP Reliability To improve PB’s reliability, for a specific SVC layer in one video chunk, M3 adds some necessary FEC packets and then sends them to the corresponding receiver. SVC layer 1 SVC layer 2 SVC layer 3 …… Optimization Dynamics 2018/12/3 IWQoS 2016
28
Reliable pseudo-broadcast with FEC
FEC coding One coding block <--> One SVC layer in a video chunk A (n, k) block, containing k source packets and (n-k) overhead packets k n-k >=k Briefly introduce the principle of FEC coding. One coding block is corresponding to one SVC layer in a video chunk. For one block, containing …, if the client receives more than any k packets, then k source packets can be successfully recovered. 2018/12/3 IWQoS 2016
29
Reliable pseudo-broadcast with FEC
FEC coding One coding block <--> One SVC layer in a video chunk A (n, k) block, containing k source packets and (n-k) overhead packets Minimum packet delivery ratio in one block ( 𝐵𝑃𝐷𝑅 𝑚𝑖𝑛 ) k n-k >=k Here, we introduce a concept of minimum packet delivery ratio in one block. For example, we divide the same piece of data into blocks with different sizes for transmission. When the block size is 6, the minimum BPDR is 50 percent. And that is 75 percent when the block size is 12. To evaluate the transmission stability of PB, we need to measure the listener’s minimum BPDR for a specific block size. 50% 75% Block size:6 Block size:12 2018/12/3 IWQoS 2016
30
Goodput measurement of PB
PB’s goodput To guarantee that all listeners can recover all video chunks FEC overhead: 𝑂=(1−𝐸( 𝐵𝑃𝐷𝑅 𝑚𝑖𝑛 ))/𝐸( 𝐵𝑃𝐷𝑅 𝑚𝑖𝑛 ) The expected goodput: 𝐸(𝐺𝑃)=𝑇𝑃∙𝐸( 𝐵𝑃𝐷𝑅 𝑚𝑖𝑛 ) To guarantee that all listeners can recover all video chunks, we use the expected minimum BPDR to calculate FEC overhead, and then measure TCP throughput to get the expected goodput. Here, we just consider the listener whose RSSI is higher than the receiver’s RSSI. 2018/12/3 IWQoS 2016
31
Goodput measurement of PB
PB’s goodput To guarantee that all listeners can recover all video chunks FEC overhead: 𝑂=(1−𝐸( 𝐵𝑃𝐷𝑅 𝑚𝑖𝑛 ))/𝐸( 𝐵𝑃𝐷𝑅 𝑚𝑖𝑛 ) The expected goodput: 𝐸(𝐺𝑃)=𝑇𝑃∙𝐸( 𝐵𝑃𝐷𝑅 𝑚𝑖𝑛 ) Experiment setups Send 100 blocks via TCP TCP payload: 1448 Bytes Block size: 100 packets The ratio of decodable blocks We conduct a verification test and use the ratio of decodable blocks to determine if the actual goodput achieves the expected value. We can observe the results under the red line, when the receiver’s RSSI is lower than negative 65 dBm, listeners can almost decode all blocks, namely the actual goodput is very close to the expected value. However, not all listeners can decode all blocks. 2018/12/3 IWQoS 2016
32
M3 system design C2 C1 ARQ Challenges Solutions AP …… layer1 FEC
Pseudo-broadcast Practicability FEC AP Reliability ARQ If the listener does not receive enough packets to decode the SVC base layer, in order to ensure the smooth video playback, it will send an ACK to request additional FEC packets. And M3 will unicast these FEC packets to the listener. SVC layer 1 SVC layer 2 SVC layer 3 …… Optimization Dynamics 2018/12/3 IWQoS 2016
33
M3 system design -50dBm -60dBm -70dBm -80dBm C4 C3 C2 C1 Challenges
Solutions Pseudo-broadcast Practicability FEC AP Reliability ARQ To optimize the overall video quality, M3 first creates an RSSI-based client hierarchy to clearly define the service set of each receiver. SVC layer 1 SVC layer 2 SVC layer 3 …… Client Hierarchy Optimization Dynamics 2018/12/3 IWQoS 2016
34
RSSI-based client hierarchy
A receiver’s Service Set Receiver itself All listeners whose RSSI values are higher than the receiver’s RSSI Constraints (1) 𝑆𝑆 1 contains all M3 clients (2) 𝑆𝑆 ℎ ⊃ 𝑆𝑆 ℎ+1 and 𝐺𝑃 ℎ < 𝐺𝑃 ℎ+1 The creation of client hierarchy All clients are sorted by an ascending order of RSSI The number of effective client hierarchy: 1 Level 1: 𝑅 1 →( 𝑅𝑆𝑆𝐼 1 , 𝑆𝑆 1 , 𝐺𝑃 1 ) ⋯ Level ℎ: 𝑅 ℎ →( 𝑅𝑆𝑆𝐼 ℎ , 𝑆𝑆 ℎ , 𝐺𝑃 ℎ ) For a receiver’s service set, consisting of the receiver itself and all listeners whose RSSI values are higher than the receiver’s RSSI To create the client hierarchy, all clients are sorted by an ascending order of RSSI And then we can use two constraints to get only one effective client hierarchy. 2018/12/3 IWQoS 2016
35
M3 system design -50dBm -60dBm -70dBm -80dBm C4 C3 C2 C1 SS1 SS4 SS2
Challenges Solutions SS3 Pseudo-broadcast Practicability FEC AP Reliability ARQ For an effective client hierarchy, we can find the optimal map between SVC layers and receivers. SVC layer 1 SVC layer 2 SVC layer 3 …… Client Hierarchy Optimization BIP Dynamics 2018/12/3 IWQoS 2016
36
Binary integer linear programming
Define an allocation matrix to represent the map between SVC layers and receivers The objective is to maximize The constraints are (C1) Smooth playback (C2) SVC decoding dependencies (C3) One layer can be sent only once (C4) All clients can at least receive the SVC base layer Total Video Rate We define an allocation matrix to represent the map between SVC layers and receivers, and then use binary integer linear programming to solve the optimization problem. The objective is to maximize … There are four constraints. C1 is to ensure … C2 is to ensure … C3 is to ensure that … And C4 is to guarantee that … 2018/12/3 IWQoS 2016
37
The parameters used in the algorithm
The vide rate (𝑉𝑅) of SVC layers A synthetic SVC DASH dataset with constant bitrate The video chunk is 2 seconds duration and consists of 5 layers Chunk size: (120, 300, 500, 350, 800) packets Video rate: (675, 1688, 2813, 1969, 4500) Kbps All clients’ 𝑅𝑆𝑆𝐼 Receivers’ 𝐺𝑃 Ok, let’s take a look at some parameters used in the algorithm. Including the video rate of SVC layers, all clients’ RSSI and Receivers’ goodput. Here, it is difficult to measure receivers’ TCP throughput and listeners’ minimum BPDR. 𝐺𝑃=𝑇𝑃∙ 𝐵𝑃𝐷𝑅 𝑚𝑖𝑛 2018/12/3 IWQoS 2016
38
M3 system design -50dBm -60dBm -70dBm -80dBm C4 C3 C2 C1 Challenges
Solutions layer2 FEC layer3 FEC Pseudo-broadcast layer1 FEC Practicability FEC AP Reliability ARQ In practical application scenarios, there is not enough time to conduct a comprehensive test for all clients. Therefore, we propose a simplified measurement method based on dynamic client feedback mechanism. SVC layer 1 SVC layer 2 SVC layer 3 …… Client Hierarchy Optimization BIP Dynamics Client Feedback 2018/12/3 IWQoS 2016
39
Client feedback mechanism
Two cases in real application scenarios New client successively joins the multicast group Here, we consider two cases in real application scenarios. The first case is that the newly-arriving client successively joins the multicast group every once in a while. R1 2018/12/3 IWQoS 2016
40
Client feedback mechanism
Two cases in real application scenarios New client successively joins the multicast group C2 When the new client arrives, the M3 server will send a special test video chunk. R1 2018/12/3 IWQoS 2016
41
Client feedback mechanism
Two cases in real application scenarios New client successively joins the multicast group Test.ELs C2 Test.BL The base layer is still sent to the current R1 to maintain the video service without interruption. And other enhancement layers are sent to the newly arrived C2. R1 2018/12/3 IWQoS 2016
42
Client feedback mechanism
Two cases in real application scenarios New client successively joins the multicast group Test feedback: RSSI & TP C2 Test feedback: RSSI & BPDRmin After finishing the test chunk, clients return test feedbacks. The server can get C2’s TCP throughput and R1’s minimum BPDR. R1 2018/12/3 IWQoS 2016
43
Client feedback mechanism
Two cases in real application scenarios New client successively joins the multicast group R1 And then M3 can change the client hierarchy. C1 2018/12/3 IWQoS 2016
44
Client feedback mechanism
Two cases in real application scenarios New client successively joins the multicast group Bandwidth fluctuation impacted by background traffic Regular feedback: RSSI & TP R1 Regular feedback: RSSI & BPDRmin The second case is the available bandwidth fluctuation impacted by background traffic. In response, all clients send regular feedbacks and enable the server to regularly update receivers’ TCP throughput and listeners’ minimum BPDR. C1 2018/12/3 IWQoS 2016
45
Outline Background Challenges M3 system design Results Conclusion
Ok, In the following, I will present the evaluation results of M3 system. 2018/12/3 IWQoS 2016
46
M3 performance evaluation
Performance metrics Buffering time: the smoothness of video playback Total video rate: the overall video quality Experiment setups One AP and 8 clients a/n, 5GHz, channel 40 Non-overlapping channel Phy rate: (6.5~144.4) Mbps RSSI: 8 equal bins in (-46~-85) dBm Here, we choose two performance metrics: buffering time and total video rate. Our aim is to verify the smoothness of video playback and the overall video quality. We deployed a real testbed in our lab. There are one AP and 8 clients. The wireless adaptors run at a 5GHz non-overlapping channel. And the RSSI range is divided into 8 equal bins. 2018/12/3 IWQoS 2016
47
System stability when clients successively arrive
RSSI(dBm) -46 ~-50 -51 ~-55 -56 ~-60 -61 ~-65 -66 ~-70 -71 ~-75 -76 ~-80 -81 ~-85 Client C1 C2 C3 C4 C5 C6 C7 C8 Arrival Order C4, C1, C6, C2, C8, C5, C3, C7 We first use a random client arrival order to verify system stability when clients successively arrive. 2018/12/3 IWQoS 2016
48
System stability when clients successively arrive
RSSI(dBm) -46 ~-50 -51 ~-55 -56 ~-60 -61 ~-65 -66 ~-70 -71 ~-75 -76 ~-80 -81 ~-85 Client C1 C2 C3 C4 C5 C6 C7 C8 Arrival Order C4, C1, C6, C2, C8, C5, C3, C7 New client has the lowest RSSI A longer initial buffer duration When the new arrival client has the lowest RSSI, it may not receive any packets until M3 server changes SVC-layer allocation and will buffer for a little longer time. 2018/12/3 IWQoS 2016
49
System stability when clients successively arrive
RSSI(dBm) -46 ~-50 -51 ~-55 -56 ~-60 -61 ~-65 -66 ~-70 -71 ~-75 -76 ~-80 -81 ~-85 Client C1 C2 C3 C4 C5 C6 C7 C8 Arrival Order C4, C1, C6, C2, C8, C5, C3, C7 New client has the lowest RSSI A longer initial buffer duration And for any time, there is at most one client for buffering. And there is no re-buffering event occurring for all clients. The above results show that M3 can adjust the SVC-layer allocation to provide smooth video streaming for all current clients whenever the new client arrives No re-buffering event 2018/12/3 IWQoS 2016
50
System stability when clients successively arrive
TVR gradually increases Measured value is very close to theoretical value Order 1 Order 1 C4, C1, C6, C2, C8, C5, C3, C7 Order 2 C3, C1, C7, C5, C2, C6, C8, C4 We use two different random client arrival orders to show the results of total video rate, with the increase of the number of clients, TVR gradually increases and the measured value is very close to the theoretical value. This means that M3 can effectively support the dynamic expansion of video multicast group. Order 2 2018/12/3 IWQoS 2016
51
System stability when clients successively arrive
TVR gradually increases Measured value is very close to theoretical value Order 1 Order 1 C4, C1, C6, C2, C8, C5, C3, C7 Order 2 C3, C1, C7, C5, C2, C6, C8, C4 After all clients arriving, TVR of these two cases are nearly the same, to show that M3 can select proper receivers according to the specific client layout. Final TVR of these two cases are nearly the same Order 2 2018/12/3 IWQoS 2016
52
The impact of background traffic
Layout 1 C1 C2 C3 C4 C5 C6 C7 C8 Layout 1 (dBm) -46 ~-50 -51 ~-55 -56 ~-60 -60 ~-65 -66 ~-70 -76 ~-75 -81 ~-85 Layout 2 -61 -71 Here, we consider system stability impacted by background traffic. The experiment process is divided into 3 phases: the first phase has no background traffic, and background traffic with different levels of intensity are introduced in the next two phases. We choose 2 different client layouts. The main differences are those clients with worse channel qualities, whose RSSI values gradually increase. Layout 2 2018/12/3 IWQoS 2016
53
The impact of background traffic
TVR gradually decreases Measured value is very close to theoretical value Layout 1 C1 C2 C3 C4 C5 C6 C7 C8 Layout 1 (dBm) -46 ~-50 -51 ~-55 -56 ~-60 -60 ~-65 -66 ~-70 -76 ~-75 -81 ~-85 Layout 2 -61 -71 For Layout 1, with the arrival and increasing intensity of background traffic, TVR gradually decreases and the measured value is very close to the theoretical value. This means that client feedback mechanism can effectively respond to the changes of the available bandwidth. Layout 2 2018/12/3 IWQoS 2016
54
The impact of background traffic
TVR gradually decreases Measured value is very close to theoretical value Layout 1 C1 C2 C3 C4 C5 C6 C7 C8 Layout 1 (dBm) -46 ~-50 -51 ~-55 -56 ~-60 -60 ~-65 -66 ~-70 -76 ~-75 -81 ~-85 Layout 2 -61 -71 For Layout 2, TVR is always stable, because the available bandwidth can still support all SVC layers even in the third phase TVR is always stable Layout 2 2018/12/3 IWQoS 2016
55
Comparison with single-layer video multicast
M3 Layout 1 Layout 1 SVC layers: (675, 1688, 2813, 1969, 4500)Kbps AVC streams: (675, 2149, 4314, 5496, 8319)Kbps Here, I will introduce the comparison between M3 and single-layer video multicast. We use an AVC DASH dataset for single-layer video multicast. And the video qualities of 5 different AVC streams are equal to 5 different SVC layers. Here, SVC introduces 10 percent bitrate overhead for each enhancement layer. SVC introduces 10% overhead for each enhancement layer M3 Layout 2 Layout 2 2018/12/3 IWQoS 2016
56
Comparison with single-layer video multicast
Eliminating the coding overhead, M3 still improves TVR by over 200% M3 Layout 1 Layout 1 SVC layers: (675, 1688, 2813, 1969, 4500)Kbps AVC streams: (675, 2149, 4314, 5496, 8319)Kbps For Layout 1, even if eliminating the coding overhead, M3 still improves TVR by over 200%. The above results prove that, in the situation of big differences of channel qualities between all clients, M3 can fully utilize Wi-Fi’s multi-rate nature to effectively improve the overall video quality. SVC introduces 10% overhead for each enhancement layer M3 Layout 2 Layout 2 2018/12/3 IWQoS 2016
57
Outline Background Challenges M3 system design Results Conclusion
Finally, I would like to make a conclusion with this presentation 2018/12/3 IWQoS 2016
58
Conclusion M3 – a practical and reliable multi-layer video multicast solution over multi-rate Wi-Fi network A multicast approach via multi-receiver pseudo-broadcast No change to APs RSSI-based client hierarchy Use BIP to optimize the overall video quality Client feedback mechanism Compared with single-layer video multicast, M3 can improve total video rate by up to 200% In this paper, we present M3, a practical … It is a multicast … and makes no change to APs. M3 creates a RSSI-based client hierarchy and uses BIP to optimize the overall video quality. Using a client feedback mechanism, M3 can adapt to the dynamic network conditions. Compared with single … 2018/12/3 IWQoS
59
Thanks! Q&A So, that’s all for M3, and I’d like to take questions now.
Thank you. 2018/12/3 IWQoS
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.