Improved Social Utility of P2P Streaming with a VBR-Based Substream Design

.

P2P Streaming Basics (1)

Share

…

Content Provider (origin)

P2P Network

01 R.Buyya+3 "Content Delivery Networks" Springer LNEE, vol.9 (2008)M.Zhanikeev -- [email protected] -- Social Utility of P2P Streaming with VBR Substream Design -- http://tinyurl.com/kyutech131114 --- 2/18...

2/18

.

P2P Streaming Basics (2)

• P2P delivery network is not a tree 02

◦ officially: multiple complimentary trees

• each peer is normally at most 3-4 hopsaway from the content server 05

• P2P streaming is always realtime 03

◦ other content can be converted torealtime via scheduled sessions

Traffic flow

Scal

e

…

…

…

…

02 myself+0 "...Extremely Scalable and Low Demand ... P2P Streaming ...Variable Bitrate" CANDAR (2013)

03 myself+0 "Multi-Source Stream Aggegation in the Cloud" Wiley Book on Advanced Content Delivery ... Clouds (2013)

05 P.Baccichet+3 "Low-delay Peer-to-Peer Streaming using Scalable Video Coding" Packet Video (2007)

M.Zhanikeev -- [email protected] -- Social Utility of P2P Streaming with VBR Substream Design -- http://tinyurl.com/kyutech131114 --- 3/18...

3/18

.

P2P Streaming: Push vs Pull

Server Client

……

Server Client

……

Pull

Push

• a well known argument 03 10

• pull: +RTT for every piece

• pull: high throughput requires manymany remote peers

• push: connect once, use for along time

03 myself+0 "Multi-Source Stream Aggegation in the Cloud" Wiley Book on Advanced Content Delivery ... Clouds (2013)

10 Z.Li+4 "Towards Low-Redundancy Push-Pull P2P Live Streaming" QShine (2008)


4/18

.

Push/Pull = BitTorrent/Substream

• pull = BitTorrent, many existing methods 11

• push = substream method

◦ many services like Coolstreaming 07, Afreeca 08, Uusee 09, etc...◦ also referred to as sliced or layered streaming

11 C.Stais+1 "Realistic Media Streaming over BitTorrent" Future Network and Mobile Summit (2012)

07 B.Li+5 "Inside the New Coolstreaming: Principles, Measurements and Performance Implications" IEEE INFOCOM (2008)

08 K.Park+4 "An Analysis of User Dynamics in P2P Live Streaming Services" ICC (2010)

09 C.Wu+2 "Diagnosing Network-wide P2P Live Streaming Inefficiencies" IEEE INFOCOM (2009)


5/18

.

The Variable World

• you normally get wide distributions ofthroughput across peers

• some method are trying to cope by applyingvariable load to peers◦ variable-size pieces in BitTorrent 06, etc.

Traffic flow

Scal

e

…

…

…

…

06 C.Gurler+2 "Variable chunk size ... and ... window for P2P streaming of scalable video" ICIP (2012)


6/18

.

CBR, VBR, SVC• CBR andVBR are self-explanatory, the best VBR isH.264 12

• SVC: recently introduced into H.264 and well studied 13

• SVC: think of it as configurable load distribution across frames

GOPBlock

Time

Fra

me

size

GOPBlock

Time

Fra

me

size

CBRVBR

GOPBlock

Time

Fra

me

size

SVC

12 "Advanced video coding for generic audiovisual services" ITU-T Recommendation H.264 (2012)

13 R.Kusching+2 "An Evaluation of TCP-based Rate-Control ... Streaming of H.264/SVC" ACM SIGMM MMsys) (2010)


7/18

.

VBR and SVC in Practice• H.264 in VBR versus SVC single layer -- real trace from 15

• H.264 has several modes: temporal, spatial, single-layer, etc.

0 20 40 60 80 100Time sequence

0

4

8

12

Fram

e siz

e (k

b)

VBRSVC

0 20 40 60 80 100Distribution sequence

0

4

8

12

Fram

e siz

e (k

b)15 P.Seeling+2 "Network Performance Evaluation with Frame Size and Quality Traces ..." IEEE Comm. Surveys... (2004)


8/18

.

Methods: Formally...

• parameters: block size, video format

• input: peers with variable e2e throughput

• output: mapping between blocks and peers

• many possible patterns:◦ VBR with same size blocks, CBR with variable size blocks, etc.

.data grain..

.adaptive versus fixed

.peer load..

.regular of irregular


9/18

.

Methods: Specifically

BLOCK

GOPFRAME

Content

Fixed Regular

FixedIrregular

AdaptiveIrregular

This method

FRE AIR

FIRBLOCK

GOP

FRAME


10/18

.

Raw Tests: Simple AIR

Parent A Client

Parent B

Connection close detected

Close connection Pick a better candidate

Bigger frames (GOP pos x)

Periodic check/update

Bigger frames (GOP pos x) Smaller frames (GOP pos y)

Re -order parents

Only for changed

frames/parents/GOP pos

Failed to receive

• AIR: frame position inGOP per peer 02

• traditionaldynamics: re-electslowest parent/child

• AIR dynamics:re-assign GOP positionson failure

• features: only 1-2parents with large throughput,others are small



11/18

.

Raw Tests: AIR Methods Work!

0.1 0.15 0.2 0.25 0.3 0.35VBR shape

0

0.2

0.4

0.6

0.8

1

VBR

free

ze p

roba

bilit

y

0 0.12 0.24 0.36SVC shape

0

0.2

0.4

0.6

0.8

1

SVC

free

ze p

roba

bilit

y

0 0.5 1 1.5CBR Level

0

0.2

0.4

0.6

0.8

1

CBR

free

ze p

roba

bilit

y

0.001:0.0050.005:0.010.01:0.050.05:0.1Lower drop probability:



12/18

.

Utility: Formally..• load distribution across substreams p

{S}k, with your downstream capacity C

• throughput/rate p{R}kand lag p

{L}kprob. distributions across k peers

• with F() meaning function of, the quality if your substream design for anymethod is:

Q = F(CS, p{R}k, p{L}k

), (1)

with the obvious rate ceiling of∑i=1..k

Ri ≤ S ≤ C. (2)

• social utility in this context is the response of your design to churn• in practice: two (evaluable scalarly) situations A and B for which utility is:

U =QB − QA∣∣B− A

∣∣ . (3)


13/18

.

Utility: Optimization

• obvious subject to optimization

maximize∑

i∈people

∑j∈time

Ui,j (4)

subject to∑

i∈peopleRi ≤ S ≤ C, (5)

∑i∈people

∑j∈time

Li,j ≤ 0. (6)


14/18

.

Utility: Simulation Models• crowd distributions p

{R}k(throughputs)

{y = (eax)−1

}k

◦ setup: a ∈ (0.03, 0.05, 0.07, 0.1, 0.2, 0.3, 0.5, 0.9) and x ∈ (1..100), same toframe size in VBR/SVC

• each self has 10 peers, churn is emulated by dropping up to 5 peers randomly(replaced with new)

• methods:

1. AIR: responds to churn by changing frame size distribution to mimic p{R}k

2. FIR.VBR: reassigns frames to substreams in descending order (both in rates andframe sizes) 02

3. FIR.GOP: responds to churn by repacking blocks as multiples of GOP06 to mimic p

{R}k

4. FRE: S is split into equal substreams, conventional parent re-election


06 C.Gurler+2 "Variable chunk size ... and ... window for P2P streaming of scalable video" ICIP (2012)


15/18

.

Utility: Results

-2 0 2 4Change

-3

-1.5

0

1.5

3

Utilit

y

Crowd: 0.03..0.05

-4 -2 0 2 4Change

-2

0

2

4

Utilit

y

Crowd: 0.07..0.1FREFIR.GOPFIR.VBRAIR

-4 -2 0 2 4Change

-4

-2

0

2

4

Utilit

y

Crowd: 0.2..0.3

-4 -2 0 2 4Change

-4-2024

Utilit

y

Crowd: 0.5..0.9


16/18

.

Wrapup

• VBR is fit for real peer crowds

• SVC is even better because transponding is easy

• social utility in the AIR method shows good response both in improving anddeteriorating conditions


17/18

.

That’s all, thank you ...


18/18

.

[01] R.Buyya+3 (2008)Content Delivery NetworksSpringer LNEE, vol.9

[02] myself+0 (2013)...Extremely Scalable and Low Demand ... P2P Streaming ...Variable BitrateCANDAR

[03] myself+0 (2013)Multi-Source Stream Aggegation in the CloudWiley Book on Advanced Content Delivery ... Clouds

[04] N.Capovilla+4 (2010)...Distributing Scalable Content over P2P NetworksMMEDIA

[05] P.Baccichet+3 (2007)Low-delay Peer-to-Peer Streaming using Scalable Video CodingPacket Video


18/18

.

[06] C.Gurler+2 (2012)Variable chunk size ... and ... window for P2P streaming of scalable videoICIP

[07] B.Li+5 (2008)Inside the New Coolstreaming: Principles, Measurements and PerformanceImplicationsIEEE INFOCOM

[08] K.Park+4 (2010)An Analysis of User Dynamics in P2P Live Streaming ServicesICC

[09] C.Wu+2 (2009)Diagnosing Network-wide P2P Live Streaming InefficienciesIEEE INFOCOM

[10] Z.Li+4 (2008)Towards Low-Redundancy Push-Pull P2P Live StreamingQShine


18/18

.

[11] C.Stais+1 (2012)Realistic Media Streaming over BitTorrentFuture Network and Mobile Summit

[12] (2012)Advanced video coding for generic audiovisual servicesITU-T Recommendation H.264

[13] R.Kusching+2 (2010)An Evaluation of TCP-based Rate-Control ... Streaming of H.264/SVCACM SIGMM MMsys)

[14] M.Fidler+3 (2007)Efficient Smoothing of Robust VBR Video Traffic ... Slice-Based...CCNC

[15] P.Seeling+2 (2004)Network Performance Evaluation with Frame Size and Quality Traces ...IEEE Comm. Surveys...


18/18

Improved Social Utility of P2P Streaming with a VBR-Based Substream Design

Technology

source stream

advanced content

advanced video

vbr substream

realistic

variable chunk

p2p streaming

network performance