Resource Allocation in Underprovisioned Multioverlay Live Video - - PowerPoint PPT Presentation

resource allocation in underprovisioned multioverlay live
SMART_READER_LITE
LIVE PREVIEW

Resource Allocation in Underprovisioned Multioverlay Live Video - - PowerPoint PPT Presentation

Resource Allocation in Underprovisioned Multioverlay Live Video Sharing Services Jiayi Liu 1 , Shakeel Ahmad 2 , Eliya Buyukkaya 1 , Raouf Hamzaoui 2 and Gwendal Simon 1 1 Telecom Bretagne 2 De Montfort University MMOG player video casting tool


slide-1
SLIDE 1

Resource Allocation in Underprovisioned Multioverlay Live Video Sharing Services

Jiayi Liu1, Shakeel Ahmad2, Eliya Buyukkaya1, Raouf Hamzaoui2 and Gwendal Simon1

1Telecom Bretagne 2De Montfort University

slide-2
SLIDE 2

MMOG player video casting tool

Massively Multiplayer Online Game collaborative tool

2 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-3
SLIDE 3

MMOG player video casting tool

Massively Multiplayer Online Game collaborative tool MMOG player video casting tool

Players emit user-generated videos Developed by the CNG (Community Network Game) project Integrated in "The Missing Ink" (www.missing-ink.com)

2 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-4
SLIDE 4

Xfire : online game player video casting platform

3 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-5
SLIDE 5

Challenges

Thousands of simultaneous video sources

4 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-6
SLIDE 6

Challenges

Thousands of simultaneous video sources Each video is seen by a dozen of players

4 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-7
SLIDE 7

Challenges

Thousands of simultaneous video sources Each video is seen by a dozen of players Video sources are low-capacity computers

4 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-8
SLIDE 8

Challenges

Thousands of simultaneous video sources Each video is seen by a dozen of players Video sources are low-capacity computers CDN is not cost-effective

4 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-9
SLIDE 9

Challenges

Thousands of simultaneous video sources Each video is seen by a dozen of players Video sources are low-capacity computers CDN is not cost-effective → peer-to-peer may be useful Solution : multioverlay P2P video streaming system

4 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-10
SLIDE 10

Avancement

1

Multioverlay P2P video streaming system

2

Inter-overlay bandwidth allocation problem

3

Performance evaluation

4

Conclusion

5 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-11
SLIDE 11

System in a Nutshell Management Server

s1 s2 s3

p1 p1 p2 p2 p2

A player can simultaneously watch several videos

6 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-12
SLIDE 12

System design points

Intra-overlay P2P video streaming :

Mesh-based : bandwidth fluctuation and peer dynamics

7 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-13
SLIDE 13

System design points

Intra-overlay P2P video streaming :

Mesh-based : bandwidth fluctuation and peer dynamics

Inter-overlay bandwidth allocation :

Peers allocate their uplink bandwidth Independent with intra-overlay video streaming

7 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-14
SLIDE 14

Avancement

1

Multioverlay P2P video streaming system

2

Inter-overlay bandwidth allocation problem

3

Performance evaluation

4

Conclusion

8 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-15
SLIDE 15

Bandwidth Allocation Problem

Peers watching several videos should share their uplink Problem : How to allocate bandwidth to overlays ?

9 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-16
SLIDE 16

Bandwidth Allocation Problem

Peers watching several videos should share their uplink Problem : How to allocate bandwidth to overlays ?

  • verlay1
  • verlay2
  • verlay3

p1

uplink : 9

p2

uplink : 5

p3

uplink : 8

p4

uplink : 8

9 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-17
SLIDE 17

Bandwidth Allocation Problem

Peers watching several videos should share their uplink Problem : How to allocate bandwidth to overlays ?

  • verlay1
  • verlay2
  • verlay3

p1

uplink : 9

p2

uplink : 5

p3

uplink : 8

p4

uplink : 8

peer subscribes to overlay 9 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-18
SLIDE 18

Bandwidth Allocation Problem

Peers watching several videos should share their uplink Problem : How to allocate bandwidth to overlays ?

  • verlay1
  • verlay2
  • verlay3

p1

uplink : 9

p2

uplink : 5

p3

uplink : 8

p4

uplink : 8 capacity : 6 demand : 18 capacity : 4 demand : 12 capacity : 8 demand : 18

9 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-19
SLIDE 19

Bandwidth Allocation Problem

Peers watching several videos should share their uplink Problem : How to allocate bandwidth to overlays ?

  • verlay1
  • verlay2
  • verlay3

p1

uplink : 9

p2

uplink : 5

p3

uplink : 8

p4

uplink : 8

6 peer allocates to overlay

capacity : 12 demand : 18 capacity : 7 demand : 12 capacity : 8 demand : 18

6 3 9 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-20
SLIDE 20

Bandwidth Allocation Problem

Peers watching several videos should share their uplink Problem : How to allocate bandwidth to overlays ?

  • verlay1
  • verlay2
  • verlay3

p1

uplink : 9

p2

uplink : 5

p3

uplink : 8

p4

uplink : 8 capacity : 18 demand : 18 capacity : 8 demand : 12 capacity : 22 demand : 18

6 2 4 3 1 2 8 4 9 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-21
SLIDE 21

Objective 1 : minimizing the waste of resources Overlay provisioning :

  • verlay capacity - overlay demand
  • verlay capacity - overlay demand ≥ 0 :
  • verprovisioned overlay (G+)
  • verlay capacity - overlay demand < 0 :

underprovisioned overlay (G−)

10 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-22
SLIDE 22

Objective 1 : minimizing the waste of resources Overlay provisioning :

  • verlay capacity - overlay demand
  • verlay capacity - overlay demand ≥ 0 :
  • verprovisioned overlay (G+)
  • verlay capacity - overlay demand < 0 :

underprovisioned overlay (G−)

Objective : minimizing total underprovisioning

  • verlay∈G− |capacity − demand|

10 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-23
SLIDE 23

Our solution : a max-flow

s1 s2 s3 p1 p2 p3 p4

uplink : 9 uplink : 5 uplink : 8 uplink : 8 capacity : 6 demand : 18 capacity : 4 demand : 12 capacity : 8 demand : 18

11 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-24
SLIDE 24

Our solution : a max-flow

s1 s2 s3 p1 p2 p3 p4 f

0/9 0/5 0/8 0/8

capacity : 6 demand : 18 capacity : 4 demand : 12 capacity : 8 demand : 18

11 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-25
SLIDE 25

Our solution : a max-flow

s1 s2 s3 p1 p2 p3 p4 f

0/9 0/5 0/8 0/8

q

0/12 0/8 0/10 11 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-26
SLIDE 26

Our solution : a max-flow

s1 s2 s3 p1 p2 p3 p4 f

9/9 5/5 8/8 8/8

q

5 1 6 4 4 8 2 12/12 8/8 10/10 11 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-27
SLIDE 27

Objective 2 : for underprovisioned system

Underprovisioned system :

  • p bandwidth <
  • verlay required bandwidth

12 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-28
SLIDE 28

Objective 2 : for underprovisioned system

Underprovisioned system :

  • p bandwidth <
  • verlay required bandwidth

Given objective 1, how to share the resource deficit ?

12 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-29
SLIDE 29

Objective 2 : for underprovisioned system

Underprovisioned system :

  • p bandwidth <
  • verlay required bandwidth

Given objective 1, how to share the resource deficit ? popularity : the most popular sources first

12 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-30
SLIDE 30

Objective 2 : for underprovisioned system

Underprovisioned system :

  • p bandwidth <
  • verlay required bandwidth

Given objective 1, how to share the resource deficit ? popularity : the most popular sources first diversity : the largest number of sources

12 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-31
SLIDE 31

Objective 2 : for underprovisioned system

Underprovisioned system :

  • p bandwidth <
  • verlay required bandwidth

Given objective 1, how to share the resource deficit ? popularity : the most popular sources first diversity : the largest number of sources

  • thers : e.g. payment : the premium sources first

12 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-32
SLIDE 32

Our solution : a min-cost max-flow

s1 s2 s3 p1 p2 p3 p4 f

0/9 0/5 0/8 0/8

q

0/12 0/8 0/10 13 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-33
SLIDE 33

Our solution : a min-cost max-flow

s1 s2 s3 p1 p2 p3 p4 f

0/9 0/5 0/8 0/8

q

0/12 0/8 0/10

add a cost function popularity : 1

n, ∀s

13 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-34
SLIDE 34

Our solution : a min-cost max-flow

s1 s2 s3 p1 p2 p3 p4 f

0/9 0/5 0/8 0/8

q

0/12 0/8 0/10

add a cost function popularity : 1

n, ∀s

diversity : 1 − 1

n, ∀s

13 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-35
SLIDE 35

Our solution : a min-cost max-flow

s1 s2 s3 p1 p2 p3 p4 f

0/9 0/5 0/8 0/8

q

0/12 0/8 0/10

add a cost function popularity : 1

n, ∀s

diversity : 1 − 1

n, ∀s

payment :

  

1, if s is premium 2, otherwise

13 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-36
SLIDE 36

Avancement

1

Multioverlay P2P video streaming system

2

Inter-overlay bandwidth allocation problem

3

Performance evaluation

4

Conclusion

14 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-37
SLIDE 37

System practicability

Peer dynamics

Peers periodically report their estimated upload bandwidth Server periodically recomputes, and sends bandwidth allocation

Light peer server communication overhead

0.8 Mbps server upload bandwidth : 100,000 peers, 1min period, 3 average watching overlays×2 bytes 0.8 Mbps server download bandwidth : 100,000 peers, 1min period, 4 bytes bandwidth and 4 bytes peer ID.

Min-cost-max-flow algorithm computation time

  • nb. peers

1,000 5,000 10,000 50,000 100,000 time (sec) 0.005 0.086 0.311 7.455 31.887

15 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-38
SLIDE 38

Simulations : models and settings

Realistic player upload bandwidth :

log-normal distribution from 256kbps to 5Mbps

16 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-39
SLIDE 39

Simulations : models and settings

Realistic player upload bandwidth :

log-normal distribution from 256kbps to 5Mbps

Real trace based MMOG player relationship :

Xfire ≃ facebook network of Smith College Pareto’80-20 : 80% videos from 20% most active players

16 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-40
SLIDE 40

Simulations : models and settings

Realistic player upload bandwidth :

log-normal distribution from 256kbps to 5Mbps

Real trace based MMOG player relationship :

Xfire ≃ facebook network of Smith College Pareto’80-20 : 80% videos from 20% most active players

Watchers’ behavior leads to a Zipf video popularity :

most popular : 330 peers ; median popularity : 7 peers

16 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-41
SLIDE 41

Results - PSNR

20 25 30 35 0.2 0.4 0.6 0.8 1 PSNR (dB) CDF of peers naive

17 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-42
SLIDE 42

Results - PSNR

20 25 30 35 0.2 0.4 0.6 0.8 1 PSNR (dB) CDF of peers naive DAC

17 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-43
SLIDE 43

Results - PSNR

20 25 30 35 0.2 0.4 0.6 0.8 1 PSNR (dB) CDF of peers naive DAC diversity-based popularity-based

17 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-44
SLIDE 44

Avancement

1

Multioverlay P2P video streaming system

2

Inter-overlay bandwidth allocation problem

3

Performance evaluation

4

Conclusion

18 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-45
SLIDE 45

Conclusion

A multioverlay P2P live video streaming system Inter-overlay bandwidth allocation problem

Minimizing global underprovisioning Sharing the resource deficit with different policies

19 / 20 Jiayi Liu Multioverlay Bandwidth Allocation

slide-46
SLIDE 46

20 / 20 Jiayi Liu Multioverlay Bandwidth Allocation