ISP-friendly Peer-assisted On-demand Streaming of Long Duration Content in BBC iPlayer

ISP-friendly Peer-assisted On-demand Streaming of Long Duration Content in BBC iPlayer

Dmytro Karamshuk1, Nishanth Sastry1, Jigna Chandaria2, Andy Secker2

1King's College London, 2BBC R&D

CD-GAIN: EPSRC project in collaboration with BBC R&D, http://bit.ly/cd-gain

This talk is about BBC iPlayer

iPlayer is popular and growing

Total monthly BBC iPlayer requests across all platforms, since 2009

Introduced in 2005 for 5K users, now covers 50% of the UK population

“iPlayer Usage Effect - A Bandwidth Explosion”http://community.plus.net/blog/2008/02/08/iplayer-usage-effect-a-bandwidth-explosion/

Content Delivery: traditional approach

Install more distributed caches Requires investments

Any alternatives?

Peer-assisted Content Delivery

Content Delivery Network

user

user user

user

useruser

average of 5K users online every sec in the first day after release

5K duplicates every second!!!

Ask users for assistance

CDNs are interested in traffic savings

Content Delivery Network

user

user user

user

useruser

Traffic gain

- volume of traffic downloaded from content delivery nodes- total volume of traffic watched by the userss

u

TT-1=G

Intuition: bigger P2P swarm yields higher gain

CDN user

1

user

2

user

3

P2P is self-scaling Bigger swarms yield higher savings

Analytical Model based on M/M/ Queues

ci= ui r iAverage number of peers online

How much we can offload to peers

Traffic gain is an exponential function of swarm’s capacity

content popularitycontent duration

G c 1 e c

- by Little`s law

How peer-assistance works in real life

… there is a number of obstacle factors they do not address

Huang, Cheng et al. [NOSSDAV’2008] Zhao, Mingchen, et al. [IMC’2013]

Balachandran et al. [IMC’2013]

Obstacle N1: ISP Friendliness

Content Delivery Network

P2P increases egress traffic, impacting bandwidth cost for ISPs, Karagiannis et al. [USENIX’2005]

Traffic savings drop from 87% to 13% if swarms are localized within ISP and city, Balachandran et al.

[IMC’2013]

Obstacle N1: ISP Friendliness

Content Delivery Network

ISP1

ISP2

P2P increase egress traffic, impacting bandwidth cost for ISPs, Karagiannis et al. [USENIX’2005]

Traffic savings drop to 13% if swarms are localized within ISP and city, Balachandran et al. [IMC’2013]

Obstacle N2: Bitrate Stratification

Content Delivery Network

ISP1

ISP2

1080p720p

480p

Obstacle N2: Bitrate Stratification

Content Delivery Network ISP2

ISP1 -768Kbps

ISP1 -1.5Mbps

1080p 720p480p

Obstacle N3: Partial Participation

Content Delivery Network ISP2

ISP1 -768Kbps

ISP1 -1.5Mbps Obstacle factors decrease size of swarms

Only 31% signed up for sharing in Akamai NetSession

Zhao, Mingchen, et al. [IMC’2013]

Users - 2.2 M

IP address – 1.3 M

Sessions - 15.9 MLondon, Sep. 2013

Understanding Obstacles with Data-driven Analysis

Dealing with ISP Friendliness

Five biggest ISPs account for over 80% of traffic

Two main bitrate formats dominate in over 70% of sessions

1080p

720p480pDealing with Bitrates

Avera

ge B

roadband S

peed

Results of the Simulations1080p

720p480p

G c 1 e c

An episode of Bad Education series, 100K views

Results of the Simulations

potential traffic gains for popular content are high despite obstacles

G c 1 e c

An episode of Bad Education series, 100K views

How about Unpopular Content?

BECAUSE swarms are small availability is less

An episode of What You Can Eat show, 1K views

G c 1 e c

Increasing Availability #1: Content Bundling

Content Bundling is not effective

Increasing Availability #2: Client Caching

Simple caching can boost savings by 23%

Savings across Content CorpusTop-5% of the content

corpus accounts for 80% of trafficMost of accesses happen in the

first day after release

Yes, it’s all about very popular content

Savings across Content Corpus

https://vimeo.com/99054874

Conclusions• with peer-assisted delivery savings grow with thedemand and size of the content despite obstacles• we give a closed form solution for traffic gains

• up to 88% of traffic can be saved

• content bundling does not work but caching canboost savings by 23%

G c 1 e c

Dmytro KaramshukKing's College London

follow me on Twitter: @karamshuk

Come to our second talk tomorrow:Room: Ballroom C

Time: 4/29/2015 8:30 — 4/29/2015 10:00

Geography Devices PricingOperators

Measurement study on 50% of the UK’s population