Tutorial - Events

Report
Part 4:
Future Directions
- Conviva Confidential -
SIGCOMM’2013 Tutorial: Hui Zhang
Key Industry Trends
1. Fierce competition from diverse industry players
2. Fragmentation on multiple key technology areas
3. Disruptive change of CDN video service and business
model
4. Availability of multiple sources of highly granular data and
big data processing technologies
5. Higher consumer and business expectations on quality
experience
6. Increasing role of cloud computing in Internet video
SIGCOMM’2013 Tutorial: Hui Zhang
Trend 1: Fierce Competition Among Diverse
Categories of Industry Players
 Emerging pure-play companies tries to dominate the new
media
 Netflix, Hulu, Youtube, Qiyi, Youku
 Media companies exploiting new ways to engage consumers
and monetize content
 HBO, ESPN, NBC, CBS, Turner, Disney
 CableTV, IPTV providers, and traditional aggregators
strengthen their business
 Verizon/RedBoxInstant, Comcast/Xfinity
 Traditional and e-commerce retailers to include video
 Amazon, Wal-Mart, Apple
 Platform technology companies to become the new OTT
cable operators
 Intel, Microsoft, Google, Apple
SIGCOMM’2013 Tutorial: Hui Zhang
Trend 2: Fragmentation of Key Video
Technology De-facto Standards
• Flash was the de facto standard
•
Browser-based experience
• Uniform development/user experience across diverse
browsers and OSes
• Four key functions:
1. rich interactive experience
2. video codec, player framework
3. streaming protocol
4. content protection
• This world is dis-integrating with Microsoft,
Google, Apple promoting alternative eco-systems
SIGCOMM’2013 Tutorial: Hui Zhang
The New Fragmented Eco-system
• Interactivity:
• Player framework:
RTMP
• Streaming protocol:
HLS
Progressive
Download
• Content protection:
SIGCOMM’2013 Tutorial: Hui Zhang
Content Publisher’s Perspective:
Multi-device Requirements
 Coverage: depending on content types
 Live sports and movie streaming require the most device
penetration (Netflix, Amazon Instant, ESPN, MLB, NFL are on
10s-100s of devices)
 Quality: viewers demand high quality across all devices
 Streamers across devices vary significantly in quality delivered
 Quality requires significant investment per device
 Analytics: comprehensive cross-platform important
 Audience, quality, and content analytics in real-time
 Content Protection: based on studio requirements
 No cross-platform content protection exists today
 Common: Encrypted HLS, REMPE, Tokenization,
PlayReady/NDS/Widevine Adobe Access DRM
SIGCOMM’2013 Tutorial: Hui Zhang
Devices and Heterogeneity Growing Steadily
Screen
PC Screen
Phone / Tablet
Screen
Devices
PC / Mac
Programming Language
Flash in browser
ActionScript
Silverlight in browser
C#
HTML5 in browser
Javascript
Native application
Objective-C
HTML5 in browser
Javascript
Video View (native
player)
Java
Flash/AIR (deprecated)
ActionScript
HTML5 in browser
Javascript
NexStreaming
Java
VisualOn
Java
…
…
Windows Phone
Silverlight
C#
Xbox 360
Native
C++ + Lua
Lakeview (Silverlight)
C#
Native
C++
Trilithium / Web MAF
Javascript
Roku
Native
Brightscript
Samsung Smart TV
Native
Javascript
Apple TV
Native
Javascript
iPhone/iPad/iPod/A
ppleTV
Android
Phones/GoogleTV
TV Screen
Video Platform
PS3
Other TVs: LG TV, Sony TV, Panasonic TV, Vizio TV
Streaming players: YouView STB, Nagra,
SIGCOMM’2013 Tutorial: Hui Zhang
High Quality on Emerging Platforms
 Achieving high quality on emerging platforms will be more
challenging than on PC
 Closed systems and early stage software platforms
 More network heterogeneity and variability with mobile
 Higher quality/bit rate demands on TV
 The drive towards high quality on emerging platforms will
be more aggressive than on PC (2-3 years vs 10 years)
 Premium content is being brought to devices at a rapid
pace
 Gap of quality is larger than on PC
SIGCOMM’2013 Tutorial: Hui Zhang
The Promise of HTML5 Video
HTML5 <video> tag is intended to be the standard
interface to play and control video in browsers
 HTML4
had generic object embed, not video specific
HTML5 video has been hindered by a lack of
agreement on
 Video
formats and codecs
 Content protection (encryption)
 Streaming protocols
 Ability to use adaptive streaming
Flash remains strong on PCs because of lack of
standardization
SIGCOMM’2013 Tutorial: Hui Zhang
Current Status of HTML5 Video
Recent agreement in video and audio formats
 MP4
for video (Firefox was last to adopt)
 AAC, MP3 for audio
Recent extensions to HTML5 show a path to
enabling adaptive streaming and content
protection
 Media
Source Extensions enable adaptive streaming
implementation in Javascript
 Encrypted Media Extensions enable basic encryption
and DRM
Browser support for extensions is not universal
 Chrome
is only browser with support today
SIGCOMM’2013 Tutorial: Hui Zhang
Next Steps in HTML5 Video
HTML5 video ecosystem must be built
 Player
frameworks
 Adaptive streaming and high quality
 Ad insertion
 Content protection
 Analytics
 Closed captioning
 Multiple language support
Timing is key
 Transition
from Flash to HTML5 in 2014 or later?
 Focus of the industry is on apps and other devices
 Ecosystem not ready for majority of publishers
SIGCOMM’2013 Tutorial: Hui Zhang
Trend 3: Disruption of CDN Video
•Business
Traditional video business model of Pure Play
CDNs
•
•
premium pricing for premium service
one stop shopping for all services, globally
• Key drivers for change
•
capital intensive business for streaming/bulk data
services
• streaming no longer commands premium pricing over
bulk transfer, and unit prices for both are dropping
… while at the same time …
• rapidly expanding video services in viewership, global
geography, higher bit rate, and longer form
• increasingly higher expectationSIGCOMM’2013
of customers
and
Tutorial: Hui Zhang
Trend: CDN Pricing
CDN pricing has decreased x1.5-2 every year for
the last 6 years
45
40
35
cents/GB
30
25
20
15
10
5
0
2006
2007
2008
2009
2010
2011
SIGCOMM’2013 Tutorial: Hui Zhang
Trend: Bitrate for Premium Content
Average bitrate has increased 20-40% every year
1800
1600
Kbps
1400
1200
1000
800
600
400
200
0
2006
2007
2008
2009
2010
2011
SIGCOMM’2013 Tutorial: Hui Zhang
Trend: Per-hour Streaming Cost
Per-hour streaming cost has decreased 15-30%
every year
10
9
cents/hour
8
7
6
5
4
3
2
1
0
2006
2007
2008
2009
2010
2011
SIGCOMM’2013 Tutorial: Hui Zhang
Unsustainable CDN Economics To Scale
With Video Requirements
Video consumes 70% of Internet capacity today
100x further growth needed by video … but
scalability is hindered because of fundamental
inefficiency in the video delivery chain
 High Content
networkPublisher
infrastructure
and transit
CDN Provider
Traditional
CDN Model
$
• Pays CDN provider
• Pays for
servers
• Pays for
transit
$
cost
ISP
Consumer
• Pays for
• Pays ISP
network
• Pays
$
infrastructure
Content
• Pays for transit
Publisher
$
SIGCOMM’2013 Tutorial: Hui Zhang
Does not
grow with
increased
video traffic
Emergence of Different CDN models
controlling different part of eco-system
Traditional CDNs
 Akamai,
Limelight, Level 3, CDNetworks, ChinaCache
 Controls
only caches
 Distributed
vs. centralized architecture
Service Provider CDNs
 Verizon,
Comcast, British Telecom, Orange Telecom, ATT
 Controls
caches, switches, wires, subscription bills
Content Publisher CDNs
 QQ,
Qiyi, Google, Netflix
 Controls
contents, caches and end-to-end pipeline
SIGCOMM’2013 Tutorial: Hui Zhang
Different Systems Constraints For
Various CDN Models
Traditional CDN
Multiple data
sources
Service Provider
CDN
Content Publisher
CDN
Server selection based on multiple data sources: viewer (quality),
server (load), network (load/congestion)
Server placement
Placement with
respect to switches.
Integration with
switches
Caching
Caching close to the
consumer vs cache
performance
Caching optimized
based on knowledge
of content
Pre-populated caches vs proxy caches vs hybrid: Pre-population for
new popular content before content is available based on historical
knowledge. Latest Game of Thrones episode viewer patterns can be
predicted based on history
Control and
coordination
Control-plane to optimize across CDNs and CDN types. Content
aware and audience aware control protocols
Data-plane
Data-plane protocols to seed CDNs with content
SIGCOMM’2013 Tutorial: Hui Zhang
ISP Participating Video Distribution
 Cost reduction, additional revenue, better subscriber
experience
Content
Publisher
Traditional
CDN Model
CDN
Exchange
• Pays CDN
provider
• Pays CDN
provider
CDN Provider
ISP
Consumer
• Pays for
servers
• Pays for
transit
• Pays for
network
infrastructure
• Pays for
transit
• Pays ISP
• Pays
Content
Publisher
• Reduced
transit cost
• Reduced
infrastructure
cost
• Reduced
Cost reduction
transit
cost
enables
video
• Pays ISP
• Pays
Content
Publisher
growth without
significant growth
of ISP
SIGCOMM’2013
Tutorial: Hui Zhang
infrastructure
Example Industry Initiatives
 Akamai Accelerated Network Partner Program





Akamai provides appliance for free
Akamai operates the CDN
Akamai services the appliance
ISP pays for bandwidth and power for the appliance
ISP gets cost benefit of caching within the network
 Netflix OpenConnect




Netflix provides appliance for free
Netflix operates OpenConnect CDN
Netflix services appliance
Traffic to appliance goes through direct peering with
Netflix
 Akamai Aura Managed CDN and Licensed CDN


Akamai provides technology and optionally operational
expertise to ISPs to operate their own CDN
These CDNs instantly become part of the Akamai
Federation
SIGCOMM’2013 Tutorial: Hui Zhang
Trend 4: Multiple Sources of Granular Data
and Big Data Processing Technologies
Fine grain data from multiple sources
Client side
 available
to content publishers, device manufacturers,
device software platform providers
Server side

available to CDN providers
Within network
 available
to ISPs
SIGCOMM’2013 Tutorial: Hui Zhang
Trend 4: Big Data Processing
Technologies and Application to Video
Big data technologies have improved significantly
to allow sophisticated processing
 Cost,
scale, real-time
Both offline analysis and run-time optimization (e.g.
control plane optimization in SIGCOMM’02)
Different industry players see different subsets of
data and perform different analysis
 Content
 CDNs:
publishers: with or without its own CDN
pure play, provider-based, publisher-based
 Consumer
 ISPs
device and software platform vendors
and network equipment vendors
SIGCOMM’2013 Tutorial: Hui Zhang
Examples of Value Extraction From Data
Analytics by Different Industry Players
Device platform Owners
CDN Equipment Vendors
Cross-publisher, cross-CDN
performance visibility
Cross-CDN performance visibility with
deep network and server-side statistics
Feedback to publishers on device
Feedback to CDN operators on
optimization, comparisons / benchmerks deployment and operation
Understand device specific usage
patterns and feedback to navigation
system
Server selection based on real-time
cross-CDN load/performance
information
Caching optimization based on devicespecific usage patterns
CDN performance comparisons /
benchmarks
SIGCOMM’2013 Tutorial: Hui Zhang
Trend 5: Growing Expectation on Quality
Experience
Business Value/Consumer Expectation
UGV
Premium
Category Leader
Quality
1% Increase in Buffering Ratio Reduces Engagement by 3 minutes in 2010
SIGCOMM’2013 Tutorial: Hui Zhang
but 6 minutes in 2012
- Conviva Confidential -
Trend 6: Growing Importance of Cloud
Processing in Internet Video
Cloud compute and storage key to end-to-end
video pipeline
 Dynamic
scaling (e.g., large sporting events)
 Enables
publishers to deliver over the Internet with
minimal infrastructure
 Geographic
reach without investment in physical
infrastructure
SIGCOMM’2013 Tutorial: Hui Zhang
Cloud Computing in Internet Video
Content preparation – Transcoding, packaging
Origin storage, publishing to CDNs
Ads – Server-side ad insertion to simplify clientside workflow complexity
Video player – video player as a service hosted in
the cloud to simplify content publisher workflows
CMS / Metadata service
Analytics
Control-plane, coordination
SIGCOMM’2013 Tutorial: Hui Zhang
Wrapping Up
• NOT all contents are the same
• Video is fundamentally different from transaction traffic
• We are at the very beginning of Internet video revolution
• video is more than 60% Internet traffic today, will be more than
90% Internet traffic in 2-3 years
• What is next?
• Premium video on big screens  zero tolerance for
poor quality: 4K + 3D video
• Mobile video
• Technical challenges
• Quality, scalability, mobility, security, usability
• Supporting diverse business models
SIGCOMM’2013 Tutorial: Hui Zhang
Opportunity for Research Communities
Internet video poses new challenges and impose
new problem formulation on our traditional areas of
interest
 Quality,
scalability, fault tolerance, security, network
control, cross layer optimization, internetworking of
different providers, interaction of technology and policy
New exciting adjacent technologies areas
 Cloud
 Big
computing
data processing
 Software-Defined
Networks
SIGCOMM’2013 Tutorial: Hui Zhang
Research Directions Discussed Earlier
User plane: video Quality of Experience metric
that captures user engagement
Data plane: adaptive video control algorithm
with web-compatible service access primitives
(e.g. HTTP chunking)
Control plane: coordinate control plane that
performs network-wide optimization
Many more …
SIGCOMM’2013 Tutorial: Hui Zhang
Example Research Questions on Cloud
Computing and Internet Video
 Video specific cloud infrastructures
 Cloud and CDN integration
 Geo distributed large data ingest and low latency response
for analytics and control-plane cloud
 Dynamic scaling algorithms based on load prediction
 Real-time big-data processing as native features of cloud
 Performance requirements

Time to publish content (especially important for news sites)

Time for player load globally

High concurrent viewers (live event)

Live transcoding latency

Availability and automatic failover
SIGCOMM’2013 Tutorial: Hui Zhang
One Example Video SDN Formulation
One entity (SDN controller) controlling end user
devices, encoders, origin servers, CDN servers,
and network elements
 How
much could quality be improved?
 How
much could efficiency/cost of infrastructure be
improved?
 What
are the most valuable data sources?
 Does
dynamic controlling of encoded bit rates at the
encoder help improve quality or is the static multi-bitrate
good enough?
SIGCOMM’2013 Tutorial: Hui Zhang
Improving CDN Performance with
Additional Data
How much improvement in quality and efficiency
can be gained by incorporating viewer experience
measurements into the server selection decision?
How much improvement in quality and efficiency
can be gained by content aware server selection?
 Content
popularity, viewing patterns, etc.
SIGCOMM’2013 Tutorial: Hui Zhang
The Return of P2P
The future video delivery is a combination of
 CDN
 Service
 Cloud
provider CDN
providers
 Peer-assisted
delivery
New problem formulations and challenges
 Control
scope is not just a single group or content, but
across an entire business entity (e.g. all video served by
HBO, Netflix, Youtube)
 Multiple
policy considerations between different industry
players
 Control
stability within entity and across entities
 Security
challenges (including resource attacks)
SIGCOMM’2013 Tutorial: Hui Zhang
Improving HTTP Streaming Protocols
Protocols need to be optimized and customized for
different environments
 Cellular
3G
 Cellular
4G
 Cable
 DSL
internet access
internet access
 FiOS
 Service
provider specifics
 Range
of bit rates (low end range for phones vs high end
range for TV screens)
How much should the algorithm know about the
network structure (i.e., cable network architecture)?
SIGCOMM’2013 Tutorial: Hui Zhang
Cross Device and Publisher Caching
Different devices use different HTTP chunking
formats (HLS, HDS, SmoothStreaming, DASH)
Different content publishers have different copies of
the same content
 For
example, consumers can access the latest episode
of Game of Thrones on HBO GO, Xfinity On Demand,
AT&T Uverse On Demand, iTunes, etc.
How can cross-device and cross-publisher caching
be enabled?
What is the efficiency / cost savings benefits of
doing this?
What are the data-plane and control-plane
requirements to do this?
SIGCOMM’2013 Tutorial: Hui Zhang

similar documents