Page: 1
Presentation Notes
Paul Southern, Pasoco Pte Ltd
Paul Southern
Speaker:
Content Manager:
Title of presentation:
Name of Event:
Location of Event:
Presentation date/time:
Length of presentation
Audience:
Thursday 15 August 2013, 4-530pm
60 (plus Q&A)
IEEE Broadcast Technology Society Singapore Chapter
Franklin at Level 11, Connexis South Tower, 1 Fusionopolis Way,
Singapore 138632
CDN
Public, non NDA. A*Star, NUS. IEEE members
Press Announcement: https://meetings.vtools.ieee.org/meeting_view/list_meeting/19605
Host: Guo Huaqun (I2R)
Page: 2
Page: 3
Page: 4
Paul Southern
• 25 years in IT, telco, media and digital
experiences
• Large MNCs and startups
• Mature and emerging markets on 6 continents
• IPTV/OTT, cloud, CDN, Asianization, UX
• Sales, strategy, CTO, business development,
development, support, Professional Services,
consulting, founder
• CTO APACcache
Page: 5
Internet megatrends
Drivers for CDN
Enterprise CDN
Transparent caching
Future of CDN
Page: 6
Internet megatrends
Page: 7
Internet megatrends 1
• Cloud: 1/3 of all data will live in the cloud by
20201, 2/3 of all workloads2
• Media: More than 90% of Internet traffic will
be video by 20151
• Mobile: 50 billion / 10:1 devices per internet
user by 20202
• Security: Cyber attacks cost companies up to
$100,000/min3
Sources: 1-Cisco, 2-Akamai 3-McAfee, all c/o Akamai
Page: 8
Internet megatrends 2
• 45% of world’s internet users in Asia1
• 11m new users every month1
• 64% of Asians purchased online2
• $1T, +21% global eCommerce revenue in 20122
• 69% of Asians watch internet video1
• 2X India online video consumption 2011-20133
Sources: 1-We are social, Oct. 2012, 2-eMarketer, Feb. 2013, 3-Comscore,
May 2013, all c/o Akamai
Page: 9
Media explosion
• Media traffic Tbps
– 2006 = 1
– 2009 = 2.6
– 2012 = 10
– 2015 est = 50
– 2020 est = hundreds
Sources: Akamai
Page: 10
Internet traffic: NA, Asiapac
Sources: Sandvine
Page: 11
Enterprise architecture
• Old
– Email access & corporate apps, using corporate PC
– Static website for information
• New
– Datacenter & dynamic website
– Corporate apps
– Branches & corporate WAN
– Intranet & VPNs
– Cloud providers (IaaS/PaaS, private/public)
– SaaS/App providers (Salesforce, Office365, etc)
– Business partners, esp SCM, CAD/CAM, media
Page: 12
Drivers for CDN
Page: 13
Drivers for CDN
• The internet – topology, investment, ownership
• Users’ expectations
Page: 14
℗
Driver #1 – The internet
• The internet is a collection of nodes and links
– National operators
– Transnational operators
– Peering points & bottlenecks
– Access control
• Commercial considerations
Page: 15
℗
Delivering internet content
• Backends (eg: CMS) to web-servers to NICs to the
internet
• Multiple operators & bandwidth costs
• Last mile
• Last meter
Page: 16
Southeast Asia
Page: 17
Example: Changi Cable Landing Station
• Station Name: Changi Cable Landing Station, or Changi CLS
• Station Owner: Singtel
• Available Backhaul Providers: Singtel, Tata, Pacnet, etc.
• Submarine Cable Systems:
– AAG
– APCN
– EAC
– TIC (Tata Indicom Cable)
– TGN-IA
– TIS (Thailand-Indonesia-Singapore)
– MIC-1 (Moratelindo International Cable-system One)
Page: 18
CDN math for 2020
• Demand (at the edge)
– 1.5B Primetime Viewers x 7 Mbps = 10,000 Tbps
– 400M Lines x 25 Mbps = 10,000 Tbps
• Supply (at the core)
– 100 Major Networks X 5 Tbps = 500 Tbps
• 500/10,000 = only 5% of demand served !
Source: Akamai, Jan 2013
Page: 19
Driver #2 – Users’ expectations
• Users’ expectations OR brands’ offerings ?
• Latency – page load, video choke, etc…
Page: 20
Latency: the business impact
• A 0.1sec increase in loading time…
= a 1% decrease in sales. (1% of Amazon’s 2012
sales = $480M)1
• A 0.5sec increase in loading time…
= a 20% decrease in traffic and ad revenues.2
• After 3sec wait…
57% of users give up, of which 80% never come
back.3
Sources: 1-Amazon c/o Ericsson, 2-Google c/o Ericsson, 2-Edgecast
Page: 21
CDN beginnings
• Pre-CDN technologies for websites: server farms,
hierarchical caching, caching proxies.
• Akamai (ex MIT) accelerates on-demand
“Breaking News” video.
• In parallel:
– Big ISPs & Hosters build their own CDNs.
– CDN network operators enter market.
• Cloud provider deploy CDN (and vice versa).
Page: 22
CDN today
• [ BTW, our browsers cache ]
• Proprietary CDNs
– Google: YouTube, Apple: iTunes , etc…
• Multi-tenant CDNs for enterprise
– Objects, downloads, video
– AJAX/JQuery CDNs (hosters)
– Hybrid / partnerships per-country
• Other CDNs
– Mobile CDNs
– Aware CDNs, eg: location, device, language
• Transcode, ad-serve
Page: 23
Enterprise CDN
Page: 24
Enterprise CDN
• Purpose:
– Enterprise offers users a better experience
• Faster page-loads, response times
• Video-quality
• Players:
– Enterprise = buyer
– CDN operator = seller (nodes and networks)
– ISPs = partner of CDN operator
Page: 25
Objects, Static & dynamic
• Static websites
– Cached objects
– Images
– Large files (eg: EXE’s, updates)
• Dynamic websites
– Not cached per se
– Route accelerated
– Compute accelerated
Page: 26
Traditional CDN
• Traditional CDN (static, DSA)
– Spread load, handle surge/spike/route failure
– UX – faster loading / playing
– Subcontract delivery
– Essential for upsell, paid/premium content
• Enabled by:
– Presence at edges
– Secret sauce, capture and redirect
– On-demand propagation
– Ability to scale, speed, per-geo, etc…
– Price (versus competitors)
– Enterprise must balance ROI
Page: 27
CDN’d objects
http://i2.cdn.turner.com/cnn/dam/assets/
130620233110-singapore-haze-6-horizontal-gallery.jpg
17KB image
evernote.s3.amazonaws.com
xxxx.ak.fbcdn.net
Page: 28
CDN’d objects
Page: 29
Traditional CDN – how it works
• Content delivery component
– Origin server and a set of edge servers (surrogates) to replicate
content
• Content distribution component
– Moves content from the origin to edge servers and ensures
consistency
• Request-routing component
– Direct user requests to edge servers
– Interact with the distribution component to keep an up-to-date
view of content
• Accounting component
– Maintains logs of client accesses and records usage of the
servers
– Assists in traffic reporting and usage-based billing
Page: 30
ChinaCache example
Government E-Commerce
Media
Finance
Games
ICPs
ISPs
Enterprises
Page: 31
Video
• Live vs On-demand
– Live = in realtime, concurrent, multicast
– On-demand = not realtime, unicast
• Big live events
– Inaugurations, weddings, sport events, chats
• Obama 2010 on CNN.com: 1.3m
• Royal wedding 2011 on Akamai: 2.9m + Others
• On-demand
– Esp viral video, breaking news
Page: 32
Plugins
Page: 33
Dynamic site acceleration (DSA)
• Oftentimes webpages are compiled (ie: html generated) when a
user does something, eg: next.
• DSA doesn’t work well for
– Highly volatile data & extreme accuracy needed, eg: seat availability
– Compute
– Search
• DSA accomplished by
– TCP optimization
– Route optimization
– Connection management
– On-the-fly compression
– SSL offload
– Pre-fetching
Page: 34
Transparent caching
Page: 35
The ISP’s dilemma #1b
Page: 36
The ISP’s dilemma #1a
Page: 37
The ISP’s dilemma #1c
Page: 38
Transparent caching
• Purpose:
– ISP to reduce cost of international bandwidth
• Players:
– Enterprise = not involved
– CDN operator = seller (nodes)
– ISPs = buyer
Page: 39
TC – performance
Page: 40
TC – performance
Page: 41
TC – What matters
• Locked-in matters: commodity versus proprietary
hardware (and software)
• Origins matter: cache the expensive origins
• All operators matter: wholesale transparent cache
(TCaaS)
• Meetme’s matter: cloud and CDN
• Rackspace matters: plethora of private CDNs
• Measurement matters: P2P will always be big
Page: 42
Future of CDN
Page: 43
CDN vs Cloud
• Cloud centralized & CDN at edge
– Complementary
– CDNs doing cloud, Clouds doing CDN
– Eg: Microsoft Azure CDN, Amazon AWS (S3, EC2,
Cloudfront)
• Many origins in cloud, eg: in AWS
– Especially good for large object, eg: AWS S3
• ChinaCache’s Orca
– Management systems in cloud (prov, billing, mtc)
– Objects from origins moved to cloud (new origin) then
cached
Page: 44
Future of CDN
• CDN standards / federated CDNs
• Off-the-shelf application-specific mini-CDNs, eg:
– Learning environments with interactive
• Similar to Go To Meeting, Webex
– Digital signage
– CSN / Content Service Netwokr
• Partnerships
– CDN operator + Cloud / Carrier / Hoster
• Proprietary networks
– Differentiation, plugins
Page: 45
Future of CDN
• Service-Oriented Architecture
– Content management based on users’ preferences
– Personalization based on data mining
– Convergence with big-data systems
• Edge compute & DSA
– Scripts, animations, DHTML, XML
– IOT / internet of things
Page: 46
Thank you
Questions?
paul@pasoco.biz

130815 - Content Delviery Networks for the IEEE Singapore Broadcast group

  • 1.
    Page: 1 Presentation Notes PaulSouthern, Pasoco Pte Ltd Paul Southern Speaker: Content Manager: Title of presentation: Name of Event: Location of Event: Presentation date/time: Length of presentation Audience: Thursday 15 August 2013, 4-530pm 60 (plus Q&A) IEEE Broadcast Technology Society Singapore Chapter Franklin at Level 11, Connexis South Tower, 1 Fusionopolis Way, Singapore 138632 CDN Public, non NDA. A*Star, NUS. IEEE members Press Announcement: https://meetings.vtools.ieee.org/meeting_view/list_meeting/19605 Host: Guo Huaqun (I2R)
  • 2.
  • 3.
  • 4.
    Page: 4 Paul Southern •25 years in IT, telco, media and digital experiences • Large MNCs and startups • Mature and emerging markets on 6 continents • IPTV/OTT, cloud, CDN, Asianization, UX • Sales, strategy, CTO, business development, development, support, Professional Services, consulting, founder • CTO APACcache
  • 5.
    Page: 5 Internet megatrends Driversfor CDN Enterprise CDN Transparent caching Future of CDN
  • 6.
  • 7.
    Page: 7 Internet megatrends1 • Cloud: 1/3 of all data will live in the cloud by 20201, 2/3 of all workloads2 • Media: More than 90% of Internet traffic will be video by 20151 • Mobile: 50 billion / 10:1 devices per internet user by 20202 • Security: Cyber attacks cost companies up to $100,000/min3 Sources: 1-Cisco, 2-Akamai 3-McAfee, all c/o Akamai
  • 8.
    Page: 8 Internet megatrends2 • 45% of world’s internet users in Asia1 • 11m new users every month1 • 64% of Asians purchased online2 • $1T, +21% global eCommerce revenue in 20122 • 69% of Asians watch internet video1 • 2X India online video consumption 2011-20133 Sources: 1-We are social, Oct. 2012, 2-eMarketer, Feb. 2013, 3-Comscore, May 2013, all c/o Akamai
  • 9.
    Page: 9 Media explosion •Media traffic Tbps – 2006 = 1 – 2009 = 2.6 – 2012 = 10 – 2015 est = 50 – 2020 est = hundreds Sources: Akamai
  • 10.
    Page: 10 Internet traffic:NA, Asiapac Sources: Sandvine
  • 11.
    Page: 11 Enterprise architecture •Old – Email access & corporate apps, using corporate PC – Static website for information • New – Datacenter & dynamic website – Corporate apps – Branches & corporate WAN – Intranet & VPNs – Cloud providers (IaaS/PaaS, private/public) – SaaS/App providers (Salesforce, Office365, etc) – Business partners, esp SCM, CAD/CAM, media
  • 12.
  • 13.
    Page: 13 Drivers forCDN • The internet – topology, investment, ownership • Users’ expectations
  • 14.
    Page: 14 ℗ Driver #1– The internet • The internet is a collection of nodes and links – National operators – Transnational operators – Peering points & bottlenecks – Access control • Commercial considerations
  • 15.
    Page: 15 ℗ Delivering internetcontent • Backends (eg: CMS) to web-servers to NICs to the internet • Multiple operators & bandwidth costs • Last mile • Last meter
  • 16.
  • 17.
    Page: 17 Example: ChangiCable Landing Station • Station Name: Changi Cable Landing Station, or Changi CLS • Station Owner: Singtel • Available Backhaul Providers: Singtel, Tata, Pacnet, etc. • Submarine Cable Systems: – AAG – APCN – EAC – TIC (Tata Indicom Cable) – TGN-IA – TIS (Thailand-Indonesia-Singapore) – MIC-1 (Moratelindo International Cable-system One)
  • 18.
    Page: 18 CDN mathfor 2020 • Demand (at the edge) – 1.5B Primetime Viewers x 7 Mbps = 10,000 Tbps – 400M Lines x 25 Mbps = 10,000 Tbps • Supply (at the core) – 100 Major Networks X 5 Tbps = 500 Tbps • 500/10,000 = only 5% of demand served ! Source: Akamai, Jan 2013
  • 19.
    Page: 19 Driver #2– Users’ expectations • Users’ expectations OR brands’ offerings ? • Latency – page load, video choke, etc…
  • 20.
    Page: 20 Latency: thebusiness impact • A 0.1sec increase in loading time… = a 1% decrease in sales. (1% of Amazon’s 2012 sales = $480M)1 • A 0.5sec increase in loading time… = a 20% decrease in traffic and ad revenues.2 • After 3sec wait… 57% of users give up, of which 80% never come back.3 Sources: 1-Amazon c/o Ericsson, 2-Google c/o Ericsson, 2-Edgecast
  • 21.
    Page: 21 CDN beginnings •Pre-CDN technologies for websites: server farms, hierarchical caching, caching proxies. • Akamai (ex MIT) accelerates on-demand “Breaking News” video. • In parallel: – Big ISPs & Hosters build their own CDNs. – CDN network operators enter market. • Cloud provider deploy CDN (and vice versa).
  • 22.
    Page: 22 CDN today •[ BTW, our browsers cache ] • Proprietary CDNs – Google: YouTube, Apple: iTunes , etc… • Multi-tenant CDNs for enterprise – Objects, downloads, video – AJAX/JQuery CDNs (hosters) – Hybrid / partnerships per-country • Other CDNs – Mobile CDNs – Aware CDNs, eg: location, device, language • Transcode, ad-serve
  • 23.
  • 24.
    Page: 24 Enterprise CDN •Purpose: – Enterprise offers users a better experience • Faster page-loads, response times • Video-quality • Players: – Enterprise = buyer – CDN operator = seller (nodes and networks) – ISPs = partner of CDN operator
  • 25.
    Page: 25 Objects, Static& dynamic • Static websites – Cached objects – Images – Large files (eg: EXE’s, updates) • Dynamic websites – Not cached per se – Route accelerated – Compute accelerated
  • 26.
    Page: 26 Traditional CDN •Traditional CDN (static, DSA) – Spread load, handle surge/spike/route failure – UX – faster loading / playing – Subcontract delivery – Essential for upsell, paid/premium content • Enabled by: – Presence at edges – Secret sauce, capture and redirect – On-demand propagation – Ability to scale, speed, per-geo, etc… – Price (versus competitors) – Enterprise must balance ROI
  • 27.
  • 28.
  • 29.
    Page: 29 Traditional CDN– how it works • Content delivery component – Origin server and a set of edge servers (surrogates) to replicate content • Content distribution component – Moves content from the origin to edge servers and ensures consistency • Request-routing component – Direct user requests to edge servers – Interact with the distribution component to keep an up-to-date view of content • Accounting component – Maintains logs of client accesses and records usage of the servers – Assists in traffic reporting and usage-based billing
  • 30.
    Page: 30 ChinaCache example GovernmentE-Commerce Media Finance Games ICPs ISPs Enterprises
  • 31.
    Page: 31 Video • Livevs On-demand – Live = in realtime, concurrent, multicast – On-demand = not realtime, unicast • Big live events – Inaugurations, weddings, sport events, chats • Obama 2010 on CNN.com: 1.3m • Royal wedding 2011 on Akamai: 2.9m + Others • On-demand – Esp viral video, breaking news
  • 32.
  • 33.
    Page: 33 Dynamic siteacceleration (DSA) • Oftentimes webpages are compiled (ie: html generated) when a user does something, eg: next. • DSA doesn’t work well for – Highly volatile data & extreme accuracy needed, eg: seat availability – Compute – Search • DSA accomplished by – TCP optimization – Route optimization – Connection management – On-the-fly compression – SSL offload – Pre-fetching
  • 34.
  • 35.
  • 36.
  • 37.
  • 38.
    Page: 38 Transparent caching •Purpose: – ISP to reduce cost of international bandwidth • Players: – Enterprise = not involved – CDN operator = seller (nodes) – ISPs = buyer
  • 39.
    Page: 39 TC –performance
  • 40.
    Page: 40 TC –performance
  • 41.
    Page: 41 TC –What matters • Locked-in matters: commodity versus proprietary hardware (and software) • Origins matter: cache the expensive origins • All operators matter: wholesale transparent cache (TCaaS) • Meetme’s matter: cloud and CDN • Rackspace matters: plethora of private CDNs • Measurement matters: P2P will always be big
  • 42.
  • 43.
    Page: 43 CDN vsCloud • Cloud centralized & CDN at edge – Complementary – CDNs doing cloud, Clouds doing CDN – Eg: Microsoft Azure CDN, Amazon AWS (S3, EC2, Cloudfront) • Many origins in cloud, eg: in AWS – Especially good for large object, eg: AWS S3 • ChinaCache’s Orca – Management systems in cloud (prov, billing, mtc) – Objects from origins moved to cloud (new origin) then cached
  • 44.
    Page: 44 Future ofCDN • CDN standards / federated CDNs • Off-the-shelf application-specific mini-CDNs, eg: – Learning environments with interactive • Similar to Go To Meeting, Webex – Digital signage – CSN / Content Service Netwokr • Partnerships – CDN operator + Cloud / Carrier / Hoster • Proprietary networks – Differentiation, plugins
  • 45.
    Page: 45 Future ofCDN • Service-Oriented Architecture – Content management based on users’ preferences – Personalization based on data mining – Convergence with big-data systems • Edge compute & DSA – Scripts, animations, DHTML, XML – IOT / internet of things
  • 46.