The University of Melbourne

Content Delivery Networks: Overlay Networks for Scaling and Enhancing in the Web
Mukaddim Pathan and Rajkumar Buyya
Grid Computing and Distributed Systems (GRIDS) Laboratory Dept. of Computer Science and Software Engineering The University of Melbourne, Australia www.gridbus.org

The City of Melbourne

Worlds most livable city Cultural diversity is the key essence Greater Melbourne population is 3.5 million Attractive tourist attractions The place to be!

The University of Melbourne

Tutorial@ADCOM 2008, Chennai, India

Founded in 1853 Australias top-ranked university in teaching and research

2

Aerial view of the University of Melbourne CDN: Overlay Networks for Scaling and Enhancing the Web

GRIDS Lab @ Melbourne University

R&D

Education Knowledge Transfer Video Showcase

The Grids Lab and Next-Generation Grid Technologies Available in YouTube !

Youngest and one of the rapidly growing research labs in our School/University:

Founded in 2002 Houses (20+) consisting of:

Research Fellows/PostDocs Research Programmers PhD candidates Honours/Masters students

Funding

University-wide collaboration:

National and International organizations Australian Research Council & DEST Many industries (Sun, StorageTek, Microsoft, IBM, Microsoft) Faculties of Science, Engineering, and Medicine

Many national and international collaborations.

Academics Industries Widely in academic and industrial users. Gridbus research team produces over 20% of our Depts research output.

Software:

Publication:

+ Community Services: e.g., IEEE TC Scalable Computing CDN: for Overlay Networks for Scaling and Enhancing the Web 3

Outline

Part I: CDN Fundamentals

Part I Content Delivery Networks: State of the art, insights and imperatives

CDN Insights CDN in Practice Akamai Case Study CDN Taxonomy

Part II: CDN Modeling and Performance

CDN Pricing CDN Performance

Part III: Advanced CDN Platforms and Applications

Mobile Dynamic CDNs Content Delivery for Community Networks CDN Internetworking
5 CDN: Overlay Networks for Scaling and Enhancing the Web

Internet Dynamics

Centralized Web!

With the rapid growth of Internet and Web

Slow

Services are competing each other for finite network and computing resources High availability and responsiveness are keys to business Web sites Large number of users are trying to simultaneously access the same Web site, causing Flash Crowd

content must traverse multiple backbones and long distances delivery may be prevented by congestion or backbone peering problems usage limited by bandwidth available at master site packet loss, congestion, and narrow pipes degrade stream quality
Source: Bruce Maggs, CCGrid 2001 Keynote 8 CDN: Overlay Networks for Scaling and Enhancing the Web

Unreliable

Not scalable

Inferior streaming quality

CDN: Overlay Networks for Scaling and Enhancing the Web

Content Delivery Network (CDN)

Abstract View

Content Delivery Networks (CDNs) emerged as a solution to Internet service degradation

Moving content to the edge of the Internet, close to end-users

Alternatives

Increased bandwidth, Web caching, Web pre-fetching

CDN advantages

Reduced server loads Distributed network traffic Reduced latency

9 CDN: Overlay Networks for Scaling and Enhancing the Web 10

Striking a balance between the costs for Web content providers & the QoS for end-users

CDN: Overlay Networks for Scaling and Enhancing the Web

CDN Architecture

CDN Supported Content/Services

Content delivery component

Static content

Origin server and a set of edge servers (surrogates) to replicate content Direct user requests to edge servers Interact with the distribution component to keep an up-to-date view of content Moves content from the origin to edge servers and ensures consistency Maintains logs of client accesses and records usage of the servers Assists in traffic reporting and usage-based billing
11 CDN: Overlay Networks for Scaling and Enhancing the Web

Static HTML pages, images, documents, software patches Audio, real-time video

Request-routing component

Streaming media

User Generated Video (UGV) Content services

Directory, e-commerce, file transfer services Large enterprises, Web service providers, media companies, and news broadcasters Media and Internet advertisement companies, data centers, ISPs, online music retailers, mobile operators, consumer electronics manufacturers, and other carrier companies Cell phone, smart phone/PDA, laptop, and desktop
12 CDN: Overlay Networks for Scaling and Enhancing the Web

Content distribution component

Sources of content

Accounting component

Customers

User interaction

CDN Insights Business Goals

Comparison with Related Systems

Features
Category

Scalability

CDNs
A collection of networked computers spanning the Internet Distribution of cache servers to the edge of the Internet Reducing Web latency during content delivery

Data Grids
Data intensive computing environment

Distributed Databases
Locally organized collection of data distributed across multiple physical locations Federation or splitting of existing database(s) Integration of existing databases and replication of database fragments in a transparent manner Integrity between multiple DBs Strong database consistency between distributed DBs Autonomous DDB sites Query processing, optimization, and Management Single authoritative entity

P2P Networks
Information retrieval network formed by ad-hoc aggregation of resources Collaboration among peers File sharing among peers

Ability to expand in order to handle new and large amounts of data, users and transactions Requires capabilities for dynamic provisioning and high quality content delivery, with low operational cost Future trend: content providers as well as end-users will pay to get high quality content Protection of content against unauthorized access and modification Requires physical, network, software, data and procedural security Future trend: reduce business interruption by combating against DDoS attacks and other malicious activities Service availability, handling possible outages and end-user experience Requires a fault-tolerant network with appropriate load balancing Future trend: distributed content location, cache consistency and routing mechanisms
13 CDN: Overlay Networks for Scaling and Enhancing the Web

Constitution

Formation of a VO of participating institutions Performance gain through data distribution by pre-staging, optimal source selection, and high speed data movement Integrity between data grid replicas Weak consistency between data grid replicas Autonomous participants Seamless analysis, collaboration, and maintenance of data across organizational and regional Boundaries Institutions who cooperate on some shared goals 14

Main goal

Security
Integrity Consistency

Integrity between caches Strong cache consistency between replicated content None Content caching

N/A Weak consistency between cached content Autonomous peers Locating or caching content, encrypting, retrieving, decrypting, and verifying content Self-interested end users/peers

Reliability, Responsiveness and Performance

Autonomy Operational Activities

Administration

Individual companies. proprietary in nature

CDN: Overlay Networks for Scaling and Enhancing the Web

CDN Evolutions

CDN Growth and Market Forecast

World Video Content Delivery Networks Market: Revenue Forecasts (2008-2012)
Revenues $ Mn

Growth Rate
1600.0 50.0% 40.0% 30.0% 20.0% 10.0% 0.0% 1400.0 1200.0 1000.0 800.0 600.0 400.0 200.0 0.0

Grow th Rate

2007

2008

2009 Year

2010

2011

2012

Market dynamics In 2000: 50 providers In 2002: 12 providers In 2004: 6 providers In 2006-07: 25 ->12 providers In 2008: 50 providers again!

Revenues $ Mn

Many CDN providers will not make it 24 months from now!! The market is expected to achieve revenues over a third of a billion in 2008 and cross the billion dollar mark in 2011
Source: Dan Rayburn, Streaming Media West, 20 May 2008 16 CDN: Overlay Networks for Scaling and Enhancing the Web

15

CDN: Overlay Networks for Scaling and Enhancing the Web

State of the Art Commercial CDNs

State of the Art Commercial CDNs (Contd)

CDN Name
Akamai www.akamai.com

Service Type
Provides CDN service, including streaming

Coverage
Covers 85% of the market. 40,000 servers in 900 networks in 70 countries. It handles 20% of total Internet traffic today Provides video streaming over consumer cable or ADSL modem connections around the globe, even over paths that have 20 router hops between server and end user Edge servers located in 72 locations around the world

Products/Solutions
Edge Platform for handling static as well as dynamic content, Edge Control for managing applications, and Network Operations Control Center (NOCC) EdgeStream video on-demand and IPTV Streaming software for video streaming

EdgeStream www.edgestream.com

Provides disrupted video streaming applications over the public Internet

Limelight Networks www.limelightnetworks.com

Provides distributed ondemand and live delivery of video, music, games and download

Limelight ContentEdge for distributed content delivery via HTTP, Limelight MediaEdge Streaming for distributed video and music delivery via streaming, and Limelight Custom CDN for custom distributed delivery solutions Global Content Caching, Extensible Rules Engine (XRE), Video On-Demand, and Live Webcasting

Mirror Image www.mirror-image.com

and so on
17 CDN: Overlay Networks for Scaling and Enhancing the Web

Provides content delivery, streaming media, Web computing and reporting services

Edge servers located in 22 countries

18

CDN: Overlay Networks for Scaling and Enhancing the Web

State of the Art Academic CDNs

CDN Name
CoDeeN www.codeen.cs.pr inceton.edu

Emerging CDN Practices

Description
CoDeeN is an academic testbed CDN built on top of PlanetLab Coral is a free P2P CDN. It is hosted on PlanetLab

Service Type
Provides caching of content and redirection of HTTP Requests Provides content replication in proportion to the contents popularity Provides replication of content, monitoring of servers and redirecting client requests to available replicas

Implementation & Testing

Implemented in C/C++ and tested on Linux (2.4/2.6) and MacOS (10.2/10.3) Implemented in C++ and tested on Linux, OpenBSD, FreeBSD, and Mac OS X

Availability
N/A

Unified content network

Integration of Content Delivery/Distribution and Content Services domains Content Service Network (CSN) as a service distribution channel for value added services On-demand content generation using Web applications based on end-user request specifications (e.g. scripts, animations, DHTML, XML) Edge computing, context-aware data caching, content replication, content blind data caching, as well as proprietary solutions (e.g. EdgeSuite, IBM WebSphere edge services) Use of XML parsing, Java serialization, reflection copy, clone copy Application Delivery Network (ADN) to host .NET and J2EE applications Capacity Provisioning Network (CPN) for trading cache capacities Content management is expected to be motivated by user preferences User personalization based on data mining
20 CDN: Overlay Networks for Scaling and Enhancing the Web

No official release yet. Coral is a Free software, licensed under GPLv2 (GNU General Public License) Globule is open source, subject to a BSD-style license and the Apache software license for the packaged Apache HTTP server

Dynamic content delivery

Coral www.coralcdn.org

Globule www.globule.org

Globule is an open source collaborative CDN

Implemented using PHP scripting, C/C++ and tested on Unix/Linux and Windows

Web services hosting

Service-Oriented Architecture

19

CDN: Overlay Networks for Scaling and Enhancing the Web

Research Directions

Outline

Load balancing and content replication in cooperative domain

Part I: CDN Fundamentals

Request locality Integration of replication and caching

CDN Insights CDN in Practice Akamai Case Study CDN Taxonomy

Deployment of market mechanisms

Economics models based on an SOA P2P approach for collaborative media streaming

Part II: CDN Modeling and Performance

Adaptive CDN for media streaming

CDN Pricing CDN Performance

Mobile dynamic CDN

High variability in demand due to user mobility

Part III: Advanced CDN Platforms and Applications

Content distribution through internetworking/peering/ brokering

CDN cooperation for global coverage with high quality performance

21 CDN: Overlay Networks for Scaling and Enhancing the Web

Mobile Dynamic CDNs Content Delivery for Community Networks CDN Internetworking
22 CDN: Overlay Networks for Scaling and Enhancing the Web

What is Akamai?

Content Delivery The Akamai Way

Research began at MIT in 1995 Company founded in August 1998 The largest CDN provider to date

What is the technology behind Akamai? How Akamai has improved the content delivery performance in the Internet? What are the design principles? How does it handle failures within its world-wide distributed network?

40000 servers in 70 countries Handles 20% of todays Internet traffic !

The market share leader (approx. 85%) Provides managed edge services to content providers Operates a global network of servers deployed at the edge of Internet Charged with highly complex, proprietary mathematical algorithms and patented technologies Partner with third-party application developers via open APIs
23 CDN: Overlay Networks for Scaling and Enhancing the Web

24

CDN: Overlay Networks for Scaling and Enhancing the Web

HTML Title Page for www.xyz.com with Embedded Objects

<html> <head> <title>Welcome to xyz.com!</title> </head> <body> <img src=http://www.xyz.com/jpgs/navbar1.jpg http://www.xyz.com/jpgs/navbar1.jpg > <img src=http://www.xyz.com/jpgs/navbar1.jpg http://www.xyz.com/jpgs/navbar1.jpg > <h1>Welcome to our Web site!</h1> <a href=page2.html>Click here to enter</a> </body> </html>
Source: Bruce Maggs, Akamai Technologies 25 CDN: Overlay Networks for Scaling and Enhancing the Web

Downloading www.xyz.com - before Akamai

DNS
1 WWW.XYZ.COM 5 3 6 2 10.10.123.8 4 7
10.10.123.8

Content Providers Web Server

User enters www.xyz.com Browser requests IP address for www.xyz.com DNS returns IP address Browser requests HTML Content provider providers web server returns HTML

Browser obtains IP addresses for hostnames listed in URLs of objects embedded on page Browser requests embedded objects Content provider providers web server returns embedded objects
Source: Bruce Maggs, Akamai Technologies 26 CDN: Overlay Networks for Scaling and Enhancing the Web

DNS Resolution
.com .net Root (InterNIC)

Downloading www.xyz.com - The Akamai Way

1 WWW.XYZ.COM 2

DNS
5 3 4

Content Providers Web Server

Local Name Server

TTL: 1 Day

6 7
TTL: 30 Minutes xyz.com DNS Servers

6
User enters www.xyz.com Browser requests IP address for www.xyz.com DNS returns IP address Browser requests HTML Content provider providers web server returns page with Akamaized URLs Browser obtains IP address of optimal Akamai server for embedded objects Browser obtains objects from optimal Akamai server
Source: Bruce Maggs, Akamai Technologies 28 CDN: Overlay Networks for Scaling and Enhancing the Web

10
Browsers Cache

1 2 9

OS
27

Source: Bruce Maggs, Akamai Technologies CDN: Overlay Networks for Scaling and Enhancing the Web

Content Delivery Using Akamai

Embedded URLs are Converted to ARLs <html> <head> <title>Welcome to xyz.com!</title> </head> ak <body> <img src= http://www.xyz.com/logos/logo.gif http://www.xyz.com/logos/logo.gif> http://www.xyz.com/jpgs/navbar1.jpg> <img src= http://www.xyz.com/jpgs/navbar1.jpg <h1>Welcome to our Web site!</h1> <a href=page2.html>Click here to enter</a> </body> </html>
Source: Bruce Maggs, Akamai Technologies 29 CDN: Overlay Networks for Scaling and Enhancing the Web

Typical Page Content

Total page Total Akamai Served 87,550 bytes 68,756 bytes

Logos 3,395 bytes

Banner Ads 16,174 bytes

Navigation Bar 9,674 bytes

Gif links 22,395 bytes Fresh Content 17,118 bytes

78%

Page Served by Akamai

30

Source: Bruce Maggs, Akamai Technologies

CDN: Overlay Networks for Scaling and Enhancing the Web

Akamai DNS Resolution

4
xyz.coms nameserver
xyz.com

DNS Maps and Time-To-Live

510.10.123.5
akamai.net

.com .net Root (InterNIC)

a212.g.akamai.net

ak.xyz.com

Map creation is based on measurements of:

Time To Live
1 day

15.15.125.6

select cluster
Akamai High-Level DNS Servers Akamai Low-Level DNS Servers

10 g.akamai.net
20.20.123.55

12a212.g.akamai.net
End User Local Name Server
30.30.123.5

11

Internet congestion System loads User demands Server status

Root
30 min.

HLDNS
30 sec.

13

Maps are constantly recalculated:

16

1 2 15

14

select servers within cluster

Browsers Cache

Every few minutes for HLDNS Every few seconds for LLDNS

LLDNS

TTL of DNS responses gets shorter further down the hierarchy

Source: Bruce Maggs, Akamai Technologies 32 CDN: Overlay Networks for Scaling and Enhancing the Web

OS

Source: Bruce Maggs, Akamai Technologies 31 CDN: Overlay Networks for Scaling and Enhancing the Web

Engineering Methodology

Design Principles

C programming language (gcc) Reliance on open-source code Automated unit and system builds and tests Staged rollout to production Independent release management Burn-in on invisible system

No single point of failure Minimal human intervention Decentralized organization Fail-over at multiple scales, redundancy Sophisticated algorithms Multiple, disjoint, reporting systems Backwards compatibility Secure and authenticated communications
Source: Bruce Maggs, Akamai Technologies 34 CDN: Overlay Networks for Scaling and Enhancing the Web

Source: Bruce Maggs, Akamai Technologies 33 CDN: Overlay Networks for Scaling and Enhancing the Web

Varieties of Failures

Real Time Web Monitor

Network

Congestion at public and private peering points Misconfigured routers and switches Inaccessible networks Memory SIMMS jumping out of their sockets Network cards screwed down but not in slot Switches configured to drop broadcasts Third-party software Problems with streaming clients/servers Personal firewalls Reporting Tools Customer-side problems Third-party measurements Distributed Denial of Services (DDoS)
35 Source: Bruce Maggs, Akamai Technologies CDN: Overlay Networks for Scaling and Enhancing the Web 36 CDN: Overlay Networks for Scaling and Enhancing the Web

Hardware

Software

Misperceptions

Network Traffic

Client Perceived Latency

Attacks

Attacks

Network Performance and Akamai Visualization

Network Metrics

Global News

Global Music

Global Retail
37 CDN: Overlay Networks for Scaling and Enhancing the Web 38 CDN: Overlay Networks for Scaling and Enhancing the Web

Keynote Results
Web Site Performance Typical Improvement with Akamai

Noon May 22

Noon May 16

Noon May 18

Noon May 19

Noon May 20

Noon May 21

Noon May 23

Noon May 24

Noon May 15

Noon May 17

Noon May 25

Web object delivered without Akamai

39 CDN: Overlay Networks for Scaling and Enhancing the Web 40

Web object delivered by Akamai

CDN: Overlay Networks for Scaling and Enhancing the Web

Outline

CDN Taxonomy
Issues for CDN Taxonomy

Part I: CDN Fundamentals

CDN Insights CDN in Practice Akamai Case Study CDN Taxonomy

CDN composition Content distribution and management Request-routing Performance measurement

Part II: CDN Modeling and Performance

Core issues

CDN Pricing CDN Performance

Part III: Advanced CDN Platforms and Applications

What is required for a harmonious CDN composition? How to perform effective content distribution and management? How to route client requests to appropriate CDN node? How to measure a CDNs performance?

Mobile Dynamic CDNs Content Delivery for Community Networks CDN Internetworking
41 CDN: Overlay Networks for Scaling and Enhancing the Web

Additional issues

How to handle wide-area failures in a CDN? How to ensure security in a wide-area CDN system? How to achieve wide-area application hosting?
42 CDN: Overlay Networks for Scaling and Enhancing the Web

Noon May 26

Noon May 27

CDN Composition
CDN composition CDN Organization Servers Relationships Interaction protocols Content/ service types

CDN Composition (Contd)

CDN organization
Overlay approach

Overlay Network Origin Replica approach approach server server Client-tosurrogate-to-origin server

Network Inter-cache elemetns interaction interaction Static Dynamic Services content content Network element- Inter-proxy Streaming to-caching proxy media Caching proxy arrays Caching proxy meshes

Application-specific servers and caches at several places in the network to handle content delivery Network components play no active role for content delivery Used by most commercial CDN provider Network components are equipped with code for identifying specific application types and for request forwarding based on predefined policies Used by some CDNs (e.g. Akamai), in addition to the overlay approach

Network approach

Servers

Origin server Replica server

44 CDN: Overlay Networks for Scaling and Enhancing the Web

43

CDN: Overlay Networks for Scaling and Enhancing the Web

CDN Composition (Contd)

CDN Composition (Contd)

Relationships

Interaction protocols

(a) Client-to-surrogate-to-origin server

(b) Network element-to-caching proxy

(c) Caching proxy arrays

45

(d) Caching proxy meshes

CDN: Overlay Networks for Scaling and Enhancing the Web

NECP: Lightweight protocol for signaling between servers and network elements for load balancing WCCP: Specifies interaction between one or more routers and one or more Web-caches CARP: Distributed caching protocol ICP: Lightweight message format used for inter-cache communication HTCP: Protocol for discovering HTTP caches, cached data, managing sets of HTTP caches and monitoring cache activity Cache Digest: Exchange protocol and data format
46 CDN: Overlay Networks for Scaling and Enhancing the Web

CDN Composition (Contd)

Content Distribution and Management

Content/service types

Static content: Frequency of change is low (e.g. static HTML pages, embedded images, executables, PDF documents, software patches, audio and/or video files) Dynamic content: Personalized for user or created on-demand (e.g. animations, scripts, and DHTML) Streaming media: Live or on-demand telecast (e.g. audio and/or video on-demand, movie files and music clips) Content services: Internet infrastructure services (e.g. directory, Web storage, file transfer, and e-commerce services)
47 CDN: Overlay Networks for Scaling and Enhancing the Web 48 CDN: Overlay Networks for Scaling and Enhancing the Web

Content Distribution and Management (Contd)

Content selection and delivery

Content Distribution and Management (Contd)

Surrogate placement strategies

Empirical-based approach: Based on some heuristics, Web site administrator empirically selects the content to be replicated to the edge servers Popularity-based approach: The most popular objects are replicated to the surrogates Object-based approach: Greedy approach to replicate content to the surrogate servers in units of objects
49

Cluster-based approach: Web content is grouped based on either correlation or access frequency and is replicated in units of content clusters
CDN: Overlay Networks for Scaling and Enhancing the Web

Center placement problem: For the placement of a given number of centers, minimize the maximum distance between a node and the nearest center Greedy and topology-informed placement: suboptimal algorithms take into account the existing information from CDN, such as workload patterns and the network topology Hotspot replica placement: Places replicas near the clients generating greatest load Tree-based replica placement: Based on the assumption that the underlying topologies are trees Scalable replica placement: Generates replicas on demand and organizes them into an application-level multicast tree Single-ISP and multi-ISP approach
50 CDN: Overlay Networks for Scaling and Enhancing the Web

Optimal placement

Content Distribution and Management (Contd)

Content Distribution and Management (Contd)

Content outsourcing

Non-cooperative pull-based

Upon cache miss, surrogate servers pull content from the origin server Used by most CDN providers Surrogate servers cooperate with each other to get the requested content in case of a cache miss Only used by Coral CDN, making use of DHTs

Cooperative pull-based

Content outsourcing (Contd)

Cooperative push-based

Pre-fetching of content to the surrogates Content is pushed to the surrogate servers from the origin, and surrogate servers cooperate to reduce replication and update cost
52 CDN: Overlay Networks for Scaling and Enhancing the Web

51

CDN: Overlay Networks for Scaling and Enhancing the Web

Content Distribution and Management (Contd)

Request-Routing

Cache organization and management

Caching techniques

Query-based: Query broadcast to other CDN servers Digest-based: Each CDN server maintains a digest of content held by the other cooperating surrogates Directory-based: A centralized server keeps content information of all the cooperating surrogates inside a cluster Hashing-based: A designated CDN server holds a content based on contents URL, IP addresses of the CDN servers, and the hashing function Semi-hashing-based: A local CDN server allocates a certain portion of its disk space to cache the most popular content for its local users and the remaining portion to cooperate with other CDN servers via a hashing function

Request-routing algorithms

Adaptive

Consider the current system condition to select a cache server for content delivery Use of heuristics to select a cache server rather than to consider current system condition
54 CDN: Overlay Networks for Scaling and Enhancing the Web

Cache Update
53 CDN: Overlay Networks for Scaling and Enhancing the Web

Non-adaptive

Request-Routing (Contd)

Performance Measurement

Request-routing mechanisms

GSLB Global awareness and smart authoritative DNS DNS-dispatching Modified DNS server HTTP redirection Information of replica server sets in HTTP headers URL rewriting Rewrite dynamically generated pages URL links Anycasting IP Anycasting and Applicationlevel anycasting
55

Performance evaluation metrics

Cache hit ratio Ratio of the number of cached content versus total content requested Reserved bandwidth Measure of the bandwidth used by the origin server Latency User perceived response time Surrogate server utilization Fraction of time during which the surrogate servers remain busy Reliability Packet-loss measurements
56 CDN: Overlay Networks for Scaling and Enhancing the Web

CDN peering Centralized directory model: Centralized directory to publish content Distributed hash table: Indexing through hash keys Flooded request model: Broadcast queries Document routing model: Authoritative peer for referral
CDN: Overlay Networks for Scaling and Enhancing the Web

Performance Measurement (Contd)

CDN Composition Taxonomy Mapping

CDN Name and Type
Akamai

Network statistics acquisition

CDN Organization
Network and overlay approach

Servers
Origin and replica servers

Relationships
Client-to-surrogateto-origin server, Network element-tocaching proxy, Interproxy

Interaction Protocols
Network elements interaction, inter-cache interaction

Content/Service Types
Static content, dynamic content, streaming media, and services (network monitoring, geographic targeting) Video streaming, video hosting services Static content, streaming media

Measurement-based study and simulations

Commercial CDNs

Edge Stream

Network approach

N/A

N/A

Real-time testbed

Network elements interaction Network elements interaction

PlanetLab CDNSim, CSIM, NS-2, OMNET++ and so on

Limelight Networks

Overlay approach

Simulators

Origin and replica servers

Client-to-surrogateto-origin server, Network element-tocaching proxy Client-to-surrogateto-origin server, Network element-tocaching proxy

Mirror Image

Network and Overlay approach

Origin and replica servers

Network elements interaction

Static content, streaming media, Web computing and reporting services

57

CDN: Overlay Networks for Scaling and Enhancing the Web

58

CDN: Overlay Networks for Scaling and Enhancing the Web

CDN Composition Taxonomy Mapping (Contd)

Content Distribution and Management Taxonomy Mapping

CDN Name Content Selection and Delivery
Content selection Full and partial-site delivery Content Clustering Users sessions based Content selection Partial-site delivery Content Clustering N/A Content selection Partial-site delivery Content Clustering N/A Content selection Partial-site delivery Content Clustering URL based Content selection Partial-site delivery Content Clustering N/A Content selection Full and partial-site delivery Content Clustering Users sessions based Content selection Full and partial-site delivery Content Clustering N/A

Surrogate Placement
Multi-ISP approach; Hotspot placement by allocating more servers to sites experiencing high load Single-ISP approach

Content Outsourcing
Non-cooperative pull-based

Cache Organization
Caching technique Intra and inter-cluster caching Cache update Update propagation On-demand Caching technique Inter-cluster caching Cache update N/A Caching technique Intra-cluster caching Cache update On-demand Caching technique Intra-cluster caching Cache update On-demand Caching technique Intra and inter-cluster caching Cache update On-demand Caching technique Intra and inter-cluster caching Cache update Cache invalidation Caching technique Intra and inter-cluster caching Cache update Adaptive cache update

CDN Name and Type

CoDeeN

CDN Organization
Overlay approach with open proxies

Servers
Origin and replica/ proxy (forward, reverse, redirector) Servers Origin and replica/ (cooperative) proxy cache servers Origin, replica, backup and/or redirector servers

Relationships
Client-tosurrogate-to-origin server, Network element-to-caching proxy, inter-proxy

Interaction Protocols
Network elements interaction, inter-cache interaction

Content/Service Types
Participating users receive better performance to most sites; only provides static content

Akamai

Edge Stream

Non-cooperative pull-based

Academic CDNs

Limelight Networks

Multi-ISP approach

Non-cooperative pull-based

Coral

Overlay approach with an underlying indexing infrastructure Overlay approach with end user nodes

Client-to-surrogateto-origin server, Network element-to-caching proxy, inter-proxy Client-to-surrogateto-origin server, Network element-to-caching proxy, inter-node

Network elements interaction, inter-cache Interaction Network elements interaction, inter-cache interaction

Most users receive better performance to participating sites; only provides static content A Web sites performance and availability is improved; provides static content and monitoring services

Mirror Image

Multi-ISP approach; Center placement following a concentrated Superstore architecture Multi-ISP approach; Topologyinformed replica placement

Non-cooperative pull-based

CoDeeN

Globule

Cooperative pullbased

Coral

Multi-ISP approach; Tree-based replica placement

Cooperative pullbased

Globule

Single-ISP approach; Best replica placement strategy is dynamically selected through regular evaluation of different strategies

Cooperative pullbased

59

CDN: Overlay Networks for Scaling and Enhancing the Web

60

CDN: Overlay Networks for Scaling and Enhancing the Web

10

Request-Routing Taxonomy Mapping

CDN Name
Akamai

Performance Measurement Taxonomy Mapping

CDN Name
Akamai

Request-routing Technique
Adaptive request-routing algorithms which takes into account server load and various network metrics Combination of DNS-based request-routing and URL rewriting

Performance Measurement
Internal measurement Network probing Traffic monitoring (proactive) External measurement Performed by a third party (Giga Information group) Internal measurement Traffic monitoring through Real Time Performance Monitoring Service (RPMS) N/A Internal measurement Network probing Traffic monitoring and reporting Internal measurement Local traffic and system monitoring Internal measurement Traffic monitoring Liveness checking of a proxy via UDP RPC Internal measurement Traffic monitoring Monitoring of server availability by the redirectors
62 CDN: Overlay Networks for Scaling and Enhancing the Web

EdgeStream Limelight Networks Mirror Image

HTTP redirection DNS-based request-routing Global Server Load Balancing (GSLB) Global awareness Smart authoritative DNS Request-routing algorithm takes into account request locality, system load, reliability, and proximity information. HTTP redirection. Request-routing algorithms with improved locality by exploiting on-the-fly network measurement and storing topology hints DNS-based request-routing Adaptive request-routing algorithms considering AS-based proximity Single-tier DNS-based request-routing EdgeStream

Limelight Networks Mirror Image

CoDeeN

Coral

CoDeeN Coral

Globule

Globule

61

CDN: Overlay Networks for Scaling and Enhancing the Web

Additional Issues for CDN Taxonomy

Failure handling

Part II Content Delivery Networks: Economic Models and Performance

Clustering, mirroring and multihoming Distributed monitoring service Intrusion detection, handling DDoS and protocol attacks Usage-based content and application delivery Akamai Edge Computing Infrastructure (ECI), active cache, ACDN
CDN: Overlay Networks for Scaling and Enhancing the Web

Security

Web application hosting

63

Outline

Pricing Dynamics

Part I: CDN Fundamentals

CDN Insights CDN in Practice Akamai Case Study CDN Taxonomy

Content providers should continue to derive value and CDN providers simultaneously should have incentives to manage and deploy infrastructure

A market mechanism is needed to price the service Simplest pricing structure which involves a content provider committing to a certain level of usage, e.g. $50TBs/month and a CDN determining a price per GB delivered based on this commitment Volume/bulk discount is usual, $0.5/GB for a traffic commitment of $40-50TBs and $0.15/GB for commitment over 100TBs Penalty if monthly commitment exceeds to induce commitment accuracy Periodic sampling of the bandwidth usage by a subscribing content provider, computing 95th percentile of usage at the end of the month and charging a price per Mbps based on it
66 CDN: Overlay Networks for Scaling and Enhancing the Web

Part II: CDN Modeling and Performance

Pricing based on aggregate usage

CDN Pricing CDN Performance

Part III: Advanced CDN Platforms and Applications

Mobile Dynamic CDNs Content Delivery for Community Networks CDN Internetworking
65 CDN: Overlay Networks for Scaling and Enhancing the Web

Percentile-based pricing

11

Pricing Dynamics (Contd)

CDN Pricing Models

Congestion pricing in networks

Determining a CDNs optimal pricing policy

Focus is on the interaction between pricing and QoS by studying the trade-off between congestion cost and capacity cost Pricing plays a key role to achieve desired QoS, when capacity can not be easily increased

Determine a content providers expected surplus from self-provisioning and that from delivering content through a CDN

Content provider chooses the option with higher expected surplus

Increase in pricing encourages users to shape their traffic and control the demand for network services, which in turn reduce network congestions Congestion pricing research do not readily transfer to CDN domain

Based on the content providers subscription decision, determine the CDNs optimal pricing policy, which maximizes expected profit Expected surplus, Uself = V - C(I) - c*L(I)

Self provisioning by content provider

Congestion reduction is not a CDNs goal

Economics of content delivery

Usage based pricing should entail volume discounts when subscribing content providers have similar traffic burstiness level

V is the benefit from responding to I requests/unit time C(I) is the cost for maintaining infrastructure, which is concave in I because of economies of scale c is the cost of each lost request and L(I) is the number of lost requests

Provisioning through a CDN

Volume discount can be sub-optimal for highly heterogeneous Bursty traffic

Profitability from a percentile-based pricing is substantially higher than usagebased billing

67 CDN: Overlay Networks for Scaling and Enhancing the Web

Expected surplus, UCDN = V + b(I)*- C0 P(X)

b(N) is the benefit per request from faster delivery through a set of N CDN servers C0 is the cost of outsourcing content delivery P(X) is the usage-based price the CDN changes the content provider
68 CDN: Overlay Networks for Scaling and Enhancing the Web

CDN Pricing Models (Contd)

Experimental Results

CDNs profit function, = CDNs expected revenue CDNs cost

First term is the sum of revenues from all content providers and the second term is modeled to be quadratic over the man volume of traffic handled by CDN A CDN does not know the outsourcing cost of individual content provider, but knows the cost distribution across content providers

Optimal prices are higher under Bursty traffic

Simulation is used to compute optimal infrastructure, associated expected surplus under self-provisioning and CDN provisioning for a given CDN price function Optimal pricing for Poisson and Bursty traffic

Price (MMPP) > Price (Mixed) > Price (Poisson) CDNs value proposition to content providers in terms of avoiding lost requests is enhanced in the presence of Bursty traffic

It leads to a price increase by a CDN

1000 content providers with arrival rate from a Pareto distribution [1000, 8000] Poisson traffic: all content providers have Poisson traffic Bursty traffic: all content providers have Bursty traffic Modeled as a Markov Modulated Poisson Process (MMPP) Mixed traffic: 500 CPs have Poisson and 500 CPs have MMPP traffic Cost of serving 233 requests/min is $804 per month Content providers outsourcing cost is drawn from a Uniform [0, 30000]
69 CDN: Overlay Networks for Scaling and Enhancing the Web

Optimal price under mixed traffic is convex and involves a volume tax rather than a volume discount

Traditional usage-based pricing is inefficient under Bursty traffic

Penalize content providers with low burstiness or content providers with high volume
70 CDN: Overlay Networks for Scaling and Enhancing the Web

Experimental Results (Contd)

Visionary Thoughts

Percentile-based pricing is more profitable for a CDN

CDNs face a non-trivial pricing problem

Focus is on quadratic price function with simplify numerical computation CDN profit is higher for mixed traffic CDN profit is almost same for Poisson and MMPP traffic

Percentile-based pricing is more profitable, but usage-based pricing is preferential as the first is perceived as form of peak pricing

It is expected that a transparent pricing plan coupled with competitive pressure will ultimately decrease the usage of percentile-based pricing in the next few years CDN pricing research has focused a monopolized market

Mean request rate is fixed Mean-based pricing can be converted to a percentile-based pricing policy

With mixed traffic, a CDN is permitted to provide volume discounts and at the same time, charge a higher price with greater traffic burstiness Drawbacks

Heterogeneous burstiness across content providers

Impact of competition on the pricing policies chosen by the CDNs could be investigated Impact of P2P and hybrid CDNs on pricing are yet to be understood

Percentile-based pricing is preferable

Another possibility is to provide content providers options to choose from usage-based or percentile-based pricing policies

It is not clear whether it is desirable Pay-as-you-go as used in Cloud and utility computing, Storage Delivery Networks, e.g. Amazon S3 and EC2, could emerge in CDN domain
72 CDN: Overlay Networks for Scaling and Enhancing the Web

Complicated billing and lack of standardization

71

Usage-based pricing is usable for similar burstiness

CDNs require to devise a longer term solution

CDN: Overlay Networks for Scaling and Enhancing the Web

12

Outline

Transport Systems

Part I: CDN Fundamentals

CDN Insights CDN in Practice Akamai Case Study CDN Taxonomy

The types of content transported by the CDN system is varied and have different QoS requirements

Part II: CDN Modeling and Performance

Transporting live streaming content has different requirements as compared to transporting dynamic Web content Aim is to transmit live streams to enhance quality and minimize distortions Optimization techniques

Transport system for live streaming

CDN Pricing CDN Performance

Overlay routing

Part III: Advanced CDN Platforms and Applications

Sophisticated encoding for lose recovery Pre-bursting for furst stream startup and pre-fetching for fast downloads

Mobile Dynamic CDNs Content Delivery for Community Networks CDN Internetworking
73 CDN: Overlay Networks for Scaling and Enhancing the Web

Transport system for Web and online applications

Aim is to optimize end user response time Non-http IP-based applications

Virtual Private Network (VPN), Voice-over-IP (VoIP)

74

Application-specific enhancement to meet end-to-end requirements Aim to find a better path through the Internet between source and destination Chooses the best overlay path (direct or indirect) Enhances end-to-end availability and end user performance

CDN: Overlay Networks for Scaling and Enhancing the Web

Empirical Evaluation of Global Overlay Routing

Experimental Setup

Aim is to study the performance and availability benefits of routing overlays on the commercial Internet

Measurement platform

A global subset of Akamai CDN is used for data collection Address the problem of picking optimum overlay paths between edge servers situated near end-users and origin servers situated in the Internet core Investigate both performance characterized by round-trip latency as well as path availability Randomly picking a small number of redundant paths (3 for Europe and North America, and 5 for Asia) achieves availability gains that approach the optimal For reasonable probing intervals (10 mins) and redundancy (2 paths), over 90% of the source-destination pairs outside Asia have latency improvements within 10% of the ideal

1100 cluster locations distributed across many different kinds of ISPs in 77 countries, 630 cities and 6 continents 15% of the clusters are located at the core, and the rest are at the edge Each cluster run a task that sends ICMP echo request of size 64bytes in every 2 minutes to each node in the core set (rate: < 10req/sec) 10 seconds timeout to deal no-response Probing each path: 3780, total: 652 million Modified standard all-pairs shortest-path algorithm is used A path is unavailable if three or more consecutive pings are lost Unavailability period: starting time for first lost ping to the ending time of last ping
76

Continent (Mnemonic) Africa (AF) Asia (AS) Central America (CA) Europe (EU) North America (NA) Oceania (OC) South America (SA)

Edge set 6 124 13 154 624 33 38

Core set 0 11 0 30 110 0 0

Data collection

Summary of outcomes

Evaluation

Paths that originate or end in Asia require 3 paths to reach the same performance level Indicating relatively infrequent network probing and measurements can provide optimal performance for almost all source-destination pairs
75 CDN: Overlay Networks for Scaling and Enhancing the Web

Overlay routing has high level of persistence over long periods of time

Measurements from China are filtered out, since their failure characteristics are remarkably different from others due to the firewall policies by Chinese government

CDN: Overlay Networks for Scaling and Enhancing the Web

Latency Reduction

Availability Gains in Overlays

Evaluation of what fraction of the poorly-connected sourcedestination pairs derive marginal, significant, or a large benefit from overlay routing Poorly-connected source-destination pairs see at least marginal benefits in over 80% of the samples

Evaluation of percentage samples where the direct path between the source and destination failed for each category The failure percentage of the direct paths ranges from 0.03% to 0.83% Asia has the poorest availability, as nine of the ten categories with the largest failure percentage have an endpoint in Asia In the presence of overlay routing, the failure percentage goes down by 0.3% - 0.5% for most categories

67% of the samples see significant or large benefits Some categories do deviate from the observation

Indirect path help mask failures of the direct path 3% of the direct paths caused 30% of the failures, and 10% of the direct paths give rise to 50% of the failures Reasons for the failures are specific chronic trouble rather than random, transient failures or short-lived congestions
78 CDN: Overlay Networks for Scaling and Enhancing the Web

A small number of paths give rise to 30% of failures

E.g. poorly-connected source-destination pairs with destinations in Africa do not derive much help from an overlay
77 CDN: Overlay Networks for Scaling and Enhancing the Web

13

Achieving Benefits in a Practical Design

Visionary Thoughts

In practice, there is a bit time delay to construct overlay paths and only a small number of indirect paths could be considered Stability of optimal paths

Overlay routing optimizations become more and more prevalent Careful path selection is very important to build overlays for performance gains CDNs should provide even higher level of performance with little or no downtime

The best paths between source to destination are quite persistent and do not change unless latency variations of all paths between them Latency variations of the paths over time cause a significant reordering of the best paths between source and destination, causing reordering of optimal paths Overlays designed for high performance show reduced availability as compared to the ideal situation This is because, better performing paths are typically constrained to share a small set of common links, leading to less path diversity and greater vulnerability

Performance gains of a predictive overlay

Due to recent e-commerce technology evolutions

CDNs should deliver content in a scalable fashion even during flash crowds, without loss of availability or performance It is required to meet novel and more stringent availability and performance requirements to support the next-generation of Internet applications

All shared links can simultaneously fail

Persistence

There is a high degree of consistency in the relative performance of alternative paths between a source-destination pair, for most pairs There is a small number of paths with high short term variations
79 CDN: Overlay Networks for Scaling and Enhancing the Web

Examples include applications such Virtual Private Networks (VPNs), Voice-over-IP (VoIP), which are highly latency sensitive
80 CDN: Overlay Networks for Scaling and Enhancing the Web

Outline
Part III Content Delivery Networks: Advanced Platforms and Applications

Part I: CDN Fundamentals

CDN Insights CDN in Practice Akamai Case Study CDN Taxonomy

Part II: CDN Modeling and Performance

CDN Pricing CDN Performance

Part III: Advanced CDN Platforms and Applications

Mobile Dynamic CDNs Content Delivery for Community Networks CDN Internetworking
82 CDN: Overlay Networks for Scaling and Enhancing the Web

Mobile Environments and CDNs

Mobile CDNs
Features Content type Typical CDN Static, dynamic, streaming Fixed Fixed Close to ISPs Medium Mobile CDN Static, dynamic, streaming Mobile Fixed Close to Base Station (BS) High

Mobile wireless networks offer a wide-range of dynamic and interactive services, such as GPS navigation information, mobile TV, vehicular traffic information, and location-oriented services
Represents a fundamentally different information medium than traditional Web Techniques to efficiently disseminate content is required to reduce latency and to minimize network traffic

Users location Surrogate location Surrogate topology Replica maintenance cost services

CDNs may offer a cost effective and scalable solution

Mobile CDNs

Overlay network of surrogate servers to deliver content in mobile environments

Content dissemination in mobile environments in an efficient and costeffective manner is challenging

Mobile node constraints (limited storage, processing power, input capability) due to the portable size of mobile devices Frequent network disconnections due to the navigational behavior of mobile users result in network fragmentation Geo-location oriented services according to user mobility Monitoring mechanism to obtain the real time status of the users
83 CDN: Overlay Networks for Scaling and Enhancing the Web

Components A set of surrogate servers, a network infrastructure, a network status monitoring mechanism, a cache manager, a content location manager and an accounting mechanism 84

Application services

User and geolocation oriented services Cooperative push-based scheme

Content outsourcing policy

Cooperative/ uncooperative pull-based scheme

CDN: Overlay Networks for Scaling and Enhancing the Web

14

Mobile CDNs Under Centralized Wireless Infrastructure

Mobile CDNs Under Ad-Hoc Wireless Infrastructure

User communicates with a central authority

Characteristics of Mobile AdHoc Network (MANET)

For cellular network, it is a 3G and/or GSM enabled base station For Wi-Fi network, it is an IEEE802.11 enabled Access Point

Although it provides a good framework for mobile CDNs, it has limited scalability and resilience Careful and systematic provisioning is required to provide required communications coverage and user QoS
85

Self-organized and selfconfigurable with Ad-hoc and P2P connectivity Dynamic network topology due to the unpredictable random mobility pattern of participating nodes Mobile node constraints

Distinct features of Vehicular Ad-Hoc Network (VANET)

Centralized wireless network infrastructure

CDN: Overlay Networks for Scaling and Enhancing the Web

More dynamic network topology No constraint in terms of processing power of uptime duration Predictable node movement

Mobile CDNs for vehicular environment

Dissemination of traffic-related information Location-oriented services Dynamic and streaming content provisioning Distributed gaming
86 CDN: Overlay Networks for Scaling and Enhancing the Web

Implementation and Experimentation Perspective

Visionary Thoughts

Simulation testbed

ORBIT, a wireless network testbed and PlanetLab, a large scale testbed could be used for real-time experiments CDNSim has the potential to be extended for mobile CDNs simulations Scope for extension

Information dissemination in mobile wireless environment is an interesting and challenging problem

Simulation software

Since traditional CDNs do not take user mobility into account, mobile CDNs could be used for accelerating information dissemination in mobile wireless environments

Realistic mobility traces Support of wireless environments Support for mobile resource-limited nodes

IBM WebExpress, specifically, client/intercept/server wireless computational model could be used for mobile CDNs Mobile computing and mobile networking are still evolving

Research issues to be addressed

Content placement technique

Taking user demand pattern and variation into account, future demand could be predicted A scalable and robust approach is required to alleviate the high sensitivity to delays Improvement is possible by CDN servers using state-of-the-art compression techniques
87 CDN: Overlay Networks for Scaling and Enhancing the Web

Disseminating dynamic content

There are lots of potential for further research Mobile CDNs research should focus on minimizing the communication latency during mobile users Web access and reducing the network traffic over wireless links
An industry specific rise for mobile CDNs could be predicted
88 CDN: Overlay Networks for Scaling and Enhancing the Web

Disseminating mobile streaming media

Only one mobile CDN, Ortiva Wireless, is existing

Outline

Content Delivery and Community Networks

Part I: CDN Fundamentals

CDN Insights CDN in Practice Akamai Case Study CDN Taxonomy

A network community is a group of people whose communication and collaboration over networks strengthens and facilitates their shared identity and goals

Content delivery paths: (a) traditional; (b) todays P2P; (c) future community networks

Part II: CDN Modeling and Performance

It is mainly driven by the common people or the average users

Community-based CDNs with users as content producers

CDN Pricing CDN Performance

Decentralization in content delivery must be combined with self-configuring, self-organizing, self-managing and self-adapting solutions at all technical layers to minimize the need for human intervention Autonomic network and overlay solutions are needed to establish and maintain an appropriate CDN over physical community networks Arbitrary and complex content services, e.g. content adaptation, transcoding, indexing, and storage, are required Dharamsala Wireless-Mesh community network came to life in February 2005, following the deregulation for outdoor use of WiFi in India
90 CDN: Overlay Networks for Scaling and Enhancing the Web

Part III: Advanced CDN Platforms and Applications

Content delivery and usage is special in community networks context

Mobile Dynamic CDNs Content Delivery for Community Networks CDN Internetworking
89 CDN: Overlay Networks for Scaling and Enhancing the Web

Spontaneous community networks model is quite successful

15

Architectural Framework

Description of Layers

Community networks are expected to play a central role in the intermediate future since they provide basic connectivity CDN implementation as overlay networks are captured through the delivery infrastructure layer in the architecture Content services networks consist of a set of services for handling multimedia content On top of the architectural framework is the QoE level, which reflects the actual experience of the end user

Community networks

Owned by individual users or groups of users sharing distributed resources in a relatively small geographical area Technology usage for connectivity: xDSL, Powerline, FTTH for fixed nodes connected to an ISP; WiMAX, MBWA, 3G/UMTS/HSDPA for nodes with wireless access to an ISP; WiFi and Bluetooth for mobile nodes and home networks Logical infrastructure built on top of the community network with the specific purpose of enabling access to content services P2P overlay network, content delivery and caching strategy for file sharing, live audio/video streaming, online gaming Infrastructure to provide a whole range of services to enhance content experience through easy navigation and personalized adaptation according to user needs Service-Oriented architecture to provide value added services as infrastructure services Assessment of end-users perceived experience Human Visual Senses (HVS) is crucial for accuracy and enhancement
92 CDN: Overlay Networks for Scaling and Enhancing the Web

Delivery infrastructures

Cross layer issues

QoE is a function of different QoS parameters at network, system and application level, without a direct translation between QoS parameters and QoE
91

relevant for the scenario of content delivery in community networks, e.g. correlation between QoS parameters at different layers, dependency between layers

Content services networks

Quality of Experience

CDN: Overlay Networks for Scaling and Enhancing the Web

Content Service Framework

Content Service Framework (Contd)

An End-to-End (E2E) infrastructure is required to provide seamless communication

Development of highly interactive applications within community networks

More general Content Networks (CNs) are evolving which integrate overlay structures and content services, e.g. content management

Designing an autonomic CDN architecture to address not only actual content delivery, but also content management orchestration, services functionalities and communications for faster production and easier access

Different services are to be placed into the overall service framework in order to make them a part of the content network infrastructure

The content service framework provides the context within which the services are placed

Service Level Agreement (SLA) should be in place to access QoS and QoE

A mechanism for cross-layer interaction is required for information exchange and retrieval from underlying layer and system components
93 CDN: Overlay Networks for Scaling and Enhancing the Web

Service description and its representation within the service registry is crucial Services can use other services through this service registry via interfaces

A content service framework allows dynamic and automatic composition of content services and opens up new business opportunities for brokerage services
94 CDN: Overlay Networks for Scaling and Enhancing the Web

Visionary Thoughts

Outline

Seamless handovers in a heterogeneous community network environment is challenging and it should be properly handled Community network should be coupled with techniques for user misbehavior detection and traffic anomaly detection Combination of P2P and classical CDNs could be beneficial

Part I: CDN Fundamentals

CDN Insights CDN in Practice Akamai Case Study CDN Taxonomy

Recent technology trends clearly indicate that neighborhood and home networks will be connected to core CDN

Part II: CDN Modeling and Performance

There is the need for supporting infrastructure services that provides more user freedom for interact and share content

CDN Pricing CDN Performance

How content is delivered, e.g. over a video streaming service What to do in case of insufficient resources, i.e. what kind of adaptation strategy should be applied What kind of incentive mechanisms should be used
95 CDN: Overlay Networks for Scaling and Enhancing the Web

Part III: Advanced CDN Platforms and Applications

Mobile Dynamic CDNs Content Delivery for Community Networks CDN Internetworking
96 CDN: Overlay Networks for Scaling and Enhancing the Web

16

Limitations of Current CDNs

Possible Solution

Commercial CDNs sign Service Level Agreements (SLAs) with customers

Objective is to provide global competitive services satisfying QoS requirements

Ad-hoc or planned peering arrangements between CDNs by leveraging existing infrastructures

Over-provision or harness external resources on-demand

A Flash Crowd or SlashDot Effect causes adverse business impact

Allows a CDN to scale-out to meet both flash crowds and anticipated increases in demand Avoids expense of running a global CDN

Very often CDNs can not provide QoS to end-users requests SLA violation to end-up costing the provider

Existing CDNs are proprietary in nature

Each has expensive closed delivery network No simultaneous usage of the resources of multiple CDNs

Achieves economics of scale, in terms of cost effectiveness and performance for both providers and endusers

CDN expenses to have led to consolidation in the market

Economy-based replication strategy for on-demand content placement Avoids SLA violation Assists to provide competitive service to catch-up the market
98 CDN: Overlay Networks for Scaling and Enhancing the Web

97

CDN: Overlay Networks for Scaling and Enhancing the Web

CDN Peering Scenario

Peering CDNs
A virtualized constellation between CDNs by leveraging existing infrastructures

(a) Expansion of a CDN provider through placing new POP

Members are multiple semi-independent autonomous entities who come together to share resources and collaborate on common goals Ad-hoc (short-term) or planned (long-term) peering arrangements

Initiator of a peering negotiation

Primary CDN

Participants

(b) Peering between CDN providers

99 CDN: Overlay Networks for Scaling and Enhancing the Web

Peering CDNs or Peers

Roles are fluid

Any participant can be acting as a primary or a peer

100 CDN: Overlay Networks for Scaling and Enhancing the Web

Peering CDNs: Abstract View

Web Server Web Server Mediator Global SR

Peering Arrangement Formation

Peering CDNs Overlay

Web Server

RRS
Web Server

PR

Web Server

` Web User2 Peering Agent (PA) Mediator PR

Request is served by local surrogate server or origin server (on cache miss)

(a) Formation of a peering arrangement

PA Request Routing System Web Server

RRS

Web Server SR Origin Server PA Request Routing System Origin Server

` Web User1

CDN 1

CDN N

User request of CDN 2 is served by external surrogate server from CDN N `

(b) A formed peering arrangement

User request of CDN 1 is served by external surrogate server from CDN 2

Web Server

CDN 2
101

Web Users

CDN: Overlay Networks for Scaling and Enhancing the Web

102

CDN: Overlay Networks for Scaling and Enhancing the Web

17

Performance Gain Through Peering

Cumulative Distribution 0.8 0.6 0.4 0.2 0 0 4000 8000 12000 16000 20000 Waiting Time (Time Units) Fair load ( = 0.6) Moderate load ( = 0.7) Heavy load ( = 0.9)
Cumulative Distribution

New Models for CDN Peering

Origin Server
CDN performance information Content provider (2) CDN selection (brokering) logic

1 0.8 0.6 0.4 0.2 0 0 4000 8000 12000 16000 20000 Waiting Time (Time Units) Fair load ( = 0.6) Moderate load ( = 0.7) Heavy load ( = 0.9)

(1) Request content User

`
(4) Content delivery
Peering Agent (PA) Selected embedded objects to be served by CDN provider

(3) Policy-based agreement

Peering Agent (PA)

Cumulative distribution of waiting time for the primary CDN

100000 Waiting Time (Time Units) 80000 Before Redirection After Redirection: ULB Baseline load, = 0.5

Replica servers
Peering Agent (PA)

CDN N

Replica servers

Forward request

60000 40000

Impact of requestredirection on the waiting time of the primary CDN

Web Server Selection Algorithm

Replica servers

CDN 1

CDN 2

20000 0 0.1 0.3 0.5 0.7 0.9 CDN: Overlay Networks for Scaling and Enhancing the Web 103 Load on Prim ary CDN

(a) Brokering-based approach

104 CDN: Overlay Networks for Scaling and Enhancing the Web

New Models for CDN Peering (Contd)

Origin Server (1) Request content specifying QoS requirements User Changed QoS requirements ` t onten (4) Clivery de Peering
Agent (PA) User defined QoS Content provider (2) Dynamic (and customized) CDN selection logic

Comparison of CDN Models

Typical CDN models Features Nature of Content Delivery Conventional CDNs Based on Web server Collaboration CDN Provider CDN-Content Provider Static Limited No No No P2P-based CDNs Based on peering and content availability Peers/Users No real Agreement (Self-interested users) N/A High No No Yes Advanced models for CDN peering Peering CDNs Based on CDN Internetworking/ Peering Primary CDN Provider CDN-Content Provider, CDN-CDN Short-term or long-term High Yes Yes No Brokeringbased Based on CDN Performance Content Provider CDNContent Provider Policy-based High Yes No, CDNs work in Parallel No QoS-driven (customized) brokering-based Based on user defined QoS (Customized) Content Provider CDN-Content Provider Dynamic High Yes No, CDNs work in parallel No

(3) Dynamic agreement with CDN provider(s) based on dynamic selection

Peering Agent (PA)

Responsibility for effective content delivery Entities in agreement

CDN N

Replica servers
Peering Agent (PA)

Replica servers Forward request

Selected embedded objects to be served by CDN provider Web Server Selection Algorithm

Agreement nature Scalability

CDN 2

CDN 1

Replica servers

Cooperation with external CDNs Cooperation between CDNs Cooperation between users

(b) QoS-driven (customized) brokering-based approach

105 CDN: Overlay Networks for Scaling and Enhancing the Web

106

CDN: Overlay Networks for Scaling and Enhancing the Web

MetaCDN.org (Broberg et al, GRIDS Lab): Content Delivery Network over Clouds

Visionary Thoughts
Challenges in implementing peering CDNs
Legal/copyright issues, global reach, consolidation in CDN market, challenges in brokering-based CDN peering, challenges in P2P-based CDN peering, and lack of incentives for cooperation Load distribution tp ensure reduced server load, less bandwidth consumption and improved content delivery performance

Technical issues for peering CDNs

Coordination mechanisms to ensure effectiveness, and to allow scalability and growth of cooperative CDNs Service and policy management for value-added services as infrastructure services, policies to support SLA negotiation and autonomous policy negotiation for time-critical short-term VO Pricing of contents and services to ensure maximum profit for providers in a competitive environment, yet maintain equilibrium of supply and demand
108 CDN: Overlay Networks for Scaling and Enhancing the Web

107

CDN: Overlay Networks for Scaling and Enhancing the Web

18

Tutorial Summary

Thanks for your attention!

R. Buyya, M. Pathan, A. Vakali (Eds.), Springer, Germany, 2008 http://www.gridbus.org/cdn/book/ More information at: http://www.gridbus.org/cdn

CDNs overcome Internet service degradation by offering infrastructure and mechanisms to delivery content and services in a scalable manner, and enhancing users Web experience This tutorial builds on academic and industrial research and developments and case studies by prominent CDN researchers around the world

It identifies potential research directions and technologies that drive future innovations It provides in-depth analysis and complete understanding of the current and future trends in the content distribution landscape

Content networking is an emerging research topic as the CDN landscape is dynamic

Ongoing changes include introduction of P2P based CDNs, integration of cloud computing, evolution of storage delivery networks...
109 CDN: Overlay Networks for Scaling and Enhancing the Web 110 CDN: Overlay Networks for Scaling and Enhancing the Web

19