62
Improving Content Delivery and Service Discovery in Networks Thesis Defense Suman Srinivasan Advisor: Professor Henning Schulzrinne Feb 5, 2016

My PhD thesis defense presentation

Embed Size (px)

Citation preview

Page 1: My PhD thesis defense presentation

Improving Content Delivery and Service Discovery in Networks

Thesis Defense

Suman Srinivasan

Advisor: Professor Henning Schulzrinne

Feb 5, 2016

Page 2: My PhD thesis defense presentation

Problem Statement: Summary

• The age of mobile and video is here

• Internet and IP architectures were not designed to optimally handle:• Large numbers of mobile devices at the edge• Large volume of multimedia traffic over TCP/IP

• We need smarter network infrastructure and architectures to • Better transport video and other multimedia

content• Over IP or other protocols• To mobile devices and other client devices

Improving Content Delivery and Service Discovery in Networks 2

Page 3: My PhD thesis defense presentation

Problem Statement I: Mobile

• Smartphone and mobile usage is growing dramatically

• Global handset data usage could exceed 20 petabytes by 2017 – TechCrunch

• In some countries, mobile traffic as a percentage of total internet traffic is already higher than 50% -MarketingLand

Improving Content Delivery and Service Discovery in Networks 3

Page 4: My PhD thesis defense presentation

Problem Statement II: Video

• Separately, the consumption of video content online is also increasing

• Research company Sandvine reports that over 70.4% of peak period traffic in North America is “real-time entertainment” (2015)• Of this, more than 54% is from just two

sources: Netflix and Youtube

• Source: Recode

Improving Content Delivery and Service Discovery in Networks 4

Page 5: My PhD thesis defense presentation

Contributions in My Thesis

1. Analyze data usage and working of core protocols (DNS, HTTP, service discovery) and video traffic

2. Implementation of new software and networking architectures

3. Designed new protocols and architectures to handle increasing data at network core

4. Architected research prototypes for content-centric networking

Improving Content Delivery and Service Discovery in Networks 5

Page 6: My PhD thesis defense presentation

In My Talk Today

• Opportunistic Networks• 7DS application suite• BonAHA library

• Content Delivery Networks• On-path CDN• ActiveCDN

• Content Centric Networks• CCNxServ

• Internet video viewing: analysis• JWPlayer’s video data set

• Will not cover• mDNS service discovery

traffic analysis and implementation improvements

• IPv6 as a content centric network protocol

• Both of these are explained in more detail in thesis

Improving Content Delivery and Service Discovery in Networks 6

Page 7: My PhD thesis defense presentation

Opportunistic Networks7DS & BonAHA

Improving Content Delivery and Service Discovery in Networks 7

Page 8: My PhD thesis defense presentation

7DS Application Suite

• Scenario: mobile devices without Internet connectivity need to exchange useful information

• Solution: 7DS application suite• 7DS = Seven Degrees of Separation

• Use local peer-to-peer wireless networks to exchange information

• Once wireless 7DS network set up, it could be used for any purpose• Getting web pages from peers; sending e-mails; file transfer

• Suman Srinivasan, Arezu Moghadam, Se Gi Hong, Henning G Schulzrinne, "7DS - Node Cooperation andInformation Exchange in Mostly Disconnected Networks", IEEE ICC 2007, Glasgow, Scotland, Jun 2007.• Arezu Moghadam, Suman Srinivasan, Henning Schulzrinne, "7DS - A Modular Platform to Develop MobileDisruption-tolerant Applications", IEEE NGMAST 2008, Wales, UK, September 2008.

Improving Content Delivery and Service Discovery in Networks 8

Page 9: My PhD thesis defense presentation

7DS: Establishing a Connection

Establishing the connection

zeroconf

Improving Content Delivery and Service Discovery in Networks 9

Page 10: My PhD thesis defense presentation

7DS: Get Data

Get available data from peers

Improving Content Delivery and Service Discovery in Networks 10

Page 11: My PhD thesis defense presentation

Design

• Peer-to-peer network set up using zeroconf service discovery• Protocol enables devices to discover each other without a DHCP server

• Proxy server serves content• If connected to Internet, functions normally• If not connected, works in “7DS mode”: connects to peers to get information

• Search engine• Allows node to search its own database for necessary information

• Multicast querying system• Allows nodes to query other peer nodes in local network• Searches can be for files or keywords

• Transport System• To forward e-mail

Improving Content Delivery and Service Discovery in Networks 11

Page 12: My PhD thesis defense presentation

Query Multicast Engine

• Used to actually exchange information among peers

• Requesting peer broadcasts a query to the network

• Responding peers reply if they have information

• Send encoded string with list of matching items

• Requesting peer retrieves suitable information

Improving Content Delivery and Service Discovery in Networks 12

Page 13: My PhD thesis defense presentation

BonAHA Library

• 7DS is only one of a new class of applications• “Ad-hoc/Mobile-P2P applications” that run in opportunistic networks

• Applications in such a network need to• Be aware of network transitions & state/metadata of other nodes• Can’t rely on client-server or stable infrastructure environments

• BonAHA library = Bonjour for Ad-Hoc Applications• Aims to be a framework that solves these problems

• Application examples• Group chat; File transfer; File synchronization; Local multiplayer games

• Suman Srinivasan, Arezu Moghadam, Henning Schulzrinne, "BonAHA: Service Discovery Framework forMobile Ad-Hoc Applications", IEEE Consumer Communications & Networking Conference 2009 (CCNC'09),Las Vegas, USA, January 2009.

Improving Content Delivery and Service Discovery in Networks 13

Page 14: My PhD thesis defense presentation

BonAHA framework

Node 2

Node 1

key21 = value21

key22 = value22

key23 = value23

key24 = value24

key11 = value11

key12 = value12

key13 = value13

key14 = value14

[2] node1.get(key13)

[1] node1.register()

[3] data =node1.fileGet(

value13);

Improving Content Delivery and Service Discovery in Networks 14

Page 15: My PhD thesis defense presentation

Code: LocationFinder

• Very simple, impractical application• Only for providing a quick code

sample

• Scenario• Two nodes meet each other

• Lack global knowledge of location

• Each can find out other’s last location information and update their own location

Improving Content Delivery and Service Discovery in Networks 15

Page 16: My PhD thesis defense presentation

Compare: Bonjour Code

Improving Content Delivery and Service Discovery in Networks 16

Page 17: My PhD thesis defense presentation

BonAHA Applications

BBS application

• Runs on iPod/iPhone

• Allows users to upload “posts”

• Other users can pick up “posts” and share their own

• Information on events, etc that they are interested in sharing

• Kiwoon Sung, Suman Srinivasan, Henning Schulzrinne, "BBS-ONE: Bulletin Board and Forum System for MobileOpportunistic Networks", IEEE WCNIS 2010, June 2010.

Improving Content Delivery and Service Discovery in Networks 17

Page 18: My PhD thesis defense presentation

Related Work

• Proem (2001)• Runs on “peerlet engine”• No public documentation of API

• JXTA Java library• Excellent for P2P• Heavyweight for our goals

• Peer2Me (2004)• File transfer on Bluetooth

• LightPeers• Sep 2007 PhD dissertation (B.

Christensen)• Similar model to BonAHA

• “Application”: Each application has its own GUID that identifies it

• “Session”: A group of nodes registered as running the application

• Code• Application app = new Application(appid);

• lpconn = new Connection(app);

• ses = lpconn.CreateSession();

• List<Session> sessions = lpconn.GetSessionList();

Improving Content Delivery and Service Discovery in Networks 18

Page 19: My PhD thesis defense presentation

Related Work

• LightPeers: differences with BonAHA• PING packet sent every second to search for peers

• In Bonjour, there is exponential backoff

• No library-daemon interface• LP “server” listens to packets

• Reimplementation of entire architecture + service discovery

• Bump app for iPhone and Android• Physically “bump” two phones to transfer data

• Actually connects to Bump server and transmits location info

• Fundamentally different architecture

Improving Content Delivery and Service Discovery in Networks 19

Page 20: My PhD thesis defense presentation

Content Delivery NetworksOn-Path CDN and ActiveCDN

Improving Content Delivery and Service Discovery in Networks 20

Page 21: My PhD thesis defense presentation

Content Delivery Networks (CDNs)

• Used by pretty much all major publishers today to deliver content to end users• Publishers offload their data to a CDN network,

which delivers content to the user

• CDNs cache content in network to get multimedia content to end user faster• Peering agreements with ISPs

• Co-located or closely located servers

• Involves some redirection

• DNS redirection: most commonly used

• The “Akamai model”

• Youtube uses HTTP redirection

• Requires setting up CDN node in advance

Wikipedia

Improving Content Delivery and Service Discovery in Networks 21

Page 22: My PhD thesis defense presentation

DNS redirection: CNN Video callflows

$ host ht.cdn.turner.com

ht.cdn.turner.com is an alias for vid.cnn.com.edgesuite.net.

vid.cnn.com.edgesuite.net is an alias for a1320.g.akamai.net.

a1320.g.akamai.net has address 96.17.77.82

a1320.g.akamai.net has address 96.17.77.83

HTTP request for

Flash player

HTTP request for

ad video

HTTP request for

content video

ht.cdn.turner.com is actually an alias for an Akamai server

Improving Content Delivery and Service Discovery in Networks 22

Page 23: My PhD thesis defense presentation

On-Path Content Distribution Networks

• Any node on path can serve content from cache• No redirection using HTTP/DNS

• Less traffic on the network; highly localized caching

• Reduced delay for end users and service providers

• Suman Srinivasan, Ivica Rimac, Volker Hilt, Moritz Steiner, Henning Schulzrinne, "Unveiling the content-centric features of TCP", IEEE ICC 2011, Kyoto, Japan, June 2011.

Improving Content Delivery and Service Discovery in Networks 23

Page 24: My PhD thesis defense presentation

On-path CDN: three possible architectures

• Initial design: came up with three possible architectures• #1. UDP signaling

• #2. UDP signaling and TCP SYN interception

• #3. TCP SYN interception, modifying SYN packet and proxy acting as a NAT

• Final implementation was of #3: a proxy that intercepted SYN and acted as a NAT

• Volker Hilt, Ivica Rimac, Suman Srinivasan,Moritz Steiner, "Method For Providing On-PathContent Distribution", January 2010

Improving Content Delivery and Service Discovery in Networks 24

Page 25: My PhD thesis defense presentation

TCP SYN packets and in-network termination

ProxyBrowser

End-user computerNetSerV router #1 NetSerV router #n

URL requestModified TCP SYN

Name resolution

Forwarded in direction of origin server

Modified TCP SYN

Has content

Sends TCP

SYN-ACK

TCP SYN-ACK

TCP ACK

TCP traffic

#C, #O, sp, 80#C, #O, sp, 80

+ URL

#C, #O, sp, 80

+ URL

#C, #N, sp, 80 + URL rewrite info

#C, #N, sp, 80

#C, #N, sp, 80

Add URL to TCP SYN

Modify #N to #O (incoming)

Modify #O to #N (outgoing)

#C = client IP

#O = origin server

#N = NetServ node

sp = source port Improving Content Delivery and Service Discovery in Networks 25

Page 26: My PhD thesis defense presentation

Screenshots : On-Path CDN Signaling

Proxy: adds host

payload to SYN

packet

On-Path CDN app:

Adds host/node info

to SYN-ACK packet

Browser set up

to use our proxy

server for its

network

connections.

Improving Content Delivery and Service Discovery in Networks 26

Page 27: My PhD thesis defense presentation

Delay Analysis – Round Trip Time (RTT)

0

100

200

300

400

500

600

1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64 67 70 73 76 79 82 85 88 91 94 97 100

CNN CNN Akamai Lucent WWW London direct

CNN server Akamai server ALU edge node ALU in-network node

Measurements performing using traceroute / scapy / apachebench

Dela

y i

n m

illise

conds

Percentile of packets

Improving Content Delivery and Service Discovery in Networks 27

Page 28: My PhD thesis defense presentation

ActiveCDN: dynamic CDN deployment using NetServ

• Similar to on-path CDN: “pop up” CDN nodes along content path• On-path CDN is user-driven• ActiveCDN is publisher driven

• Dynamic content migration• Publisher driven: sends signal to routers

• Built on NetServ• Next-generation network virtualization platform• Built at IRT lab

• Working• Users request content from publisher’s server• Publisher sends content, along with signal to

NetServ nodes• NetServ nodes on path download ActiveCDN

module, install it and prime the content• When second user requests, ActiveCDN node

caches and displays content from its node

Improving Content Delivery and Service Discovery in Networks 28

Page 29: My PhD thesis defense presentation

Related Work

• Layer 4/7 switches [Web Switch, 2011], also known as web or content switches: similar functionality, used primarily for caching repeatedly requested content. • Need to keep track of every TCP connection that they are routing. Impractical in a

large network.

• TCP interception has been enabled by some router vendors (Cisco, Juniper)• Initially introduced to allow systems to reduce TCP SYN-flooding attacks• Has also been used for a variety of other purposes, such as speeding up content

delivery on wireless networks, such as with NetBouncer [Thomas et al., 2003]

• CoralCDN [Freedman et al., 2004]: content is served from P2P nodes that join and leave the network. • Request made for http://<website.com>.nyud.net• DNS requests resolved by a Coral DNS server, checks for HTTP proxies near client• User redirected to a Coral node near the user that contains the content.

Improving Content Delivery and Service Discovery in Networks 29

Page 30: My PhD thesis defense presentation

• Content Centric Networks: content is a first class citizen, not hosts• As opposed to IP networks, which relies on hosts (IP addresses, domain names)

• CCNs usually involve “clean slate” architectures

• i3 [Kannan et al., 2004], OCALA [Joseph et al., 2006] and Data-Oriented Network Architecture (DONA) [Koponen et al., 2007]• Naming overlays that work on top of the existing Internet architectures.

• Address the issue of Internet naming and name resolution by allowing a client to request content by name, rather than using a host address.

• Content-Centric Networking (CCN) [Jacobson et al., 2009b] aims at treating content as a primitive for routing requests to the destination.• Make a request for ccnx://content ; get result from network

Related Work: Content Centric Networks

Improving Content Delivery and Service Discovery in Networks 30

Page 31: My PhD thesis defense presentation

CCNxServ: Service layer on CCNx

• CCNx handles content

• But users need services in addition to content

• CCNxServ aims to extend CCNx to add a service layer on top of CCNx

• Service modules are packaged as CCNx content

• Requested as ccnx://content+service

• Suman Srinivasan, Amandeep Singh, Dhruva Batni, Jae WooLee, Henning Schulzrinne, Volker Hilt, Gerald Kunzmann,"CCNxServ: Dynamic Service Scalability in Information-CentricNetworks", IEEE ICC 2012, Ottawa, Canada, June 2012.

Improving Content Delivery and Service Discovery in Networks 31

Page 32: My PhD thesis defense presentation

Real World Video ConsumptionAnalysis done at JWPlayer

Improving Content Delivery and Service Discovery in Networks 32

Page 33: My PhD thesis defense presentation

Real World Video Consumption

• Worked at JWPlayer from 2011 to 2013• Internet video startup based on NYC• Raised $20M Series D in Jan 2016 / TechCrunch• Used by sites such as KickStarter, The Guardian,

Washington Post, etc• “Over half a billion videos are watched on JW Player

video player every day” – JW Player Blog, Dec 2015

• Presenting summary of findings in this presentation: more in the full dissertation

• Dataset when I was working at JWPlayer (2013) • Serving close to 5 billion streams monthly from

videos embedded in 2 million domains • Engagement data on 40 million streams (JW hosting)

Improving Content Delivery and Service Discovery in Networks 33

Page 34: My PhD thesis defense presentation

JWPlayer Analytics 1.0 Architecture

• Industry presentation• At HBaseCon 2013

• How we built a real-time analytics platform at JWPlayer using Hadoop and HBase• 16 billion events a month• 156 million unique viewers• 1 billion video streams• 30 million hours of video watched

• Presentation available on my SlideShare

Improving Content Delivery and Service Discovery in Networks 34

Page 35: My PhD thesis defense presentation

Distribution of Video Play by Top Countries

• Distribution for the top 100 videos (by stream count)• By country where viewer is located

• We look at the top 10 countries of viewership, for top 100 videos

• We want to find the popularity distribution of popular videos across country• Are videos universally popular

around the world?

Improving Content Delivery and Service Discovery in Networks 35

Page 36: My PhD thesis defense presentation

Distribution of Video Play by Top Countries

• Distribution of video play across top 10 countries of viewership

• Most videos are primarily popular in one country

• Top country accounts for 80% of viewership (in median)• Doesn’t account for language

regions

• Viral popularity of content: restricted to country?

Improving Content Delivery and Service Discovery in Networks 36

Page 37: My PhD thesis defense presentation

Distribution of Video Play by Top Regions

• OK, so video popularity distribution is heavily skewed towards one country

• But how about popularity distribution within a country?

• Is it similarly skewed? Or more evenly distributed?

• We look at distribution for the top 100 videos across regions / states for the #1 country for each video.

Improving Content Delivery and Service Discovery in Networks 37

Page 38: My PhD thesis defense presentation

Distribution of Video Play by Top Regions

• When we break down data a bit further by states / regions within top country

• Distribution is a bit more even.

• Between 25%-50% of the traffic is from the top region / state

• But a bit more evenly distributed across other regions / states

• So within a country: distribution is more evenly distributed across states / regions

Improving Content Delivery and Service Discovery in Networks 38

Page 39: My PhD thesis defense presentation

Other Analysis

• Video popularity• Popularity distribution based on time of day / week• Popularity distribution based on time of day / week by country

• Video length• Popularity based on length of video• Video length viewed across different countries, devices, bandwidth quality

• Popularity distribution across domains (Zip-f? Power law? Linear?)

• Usage across desktop, mobile and tablet devices

• State transitions in engagement data• How many people watch a video all the way to the end?• How many people keep skipping portions of the video?

Improving Content Delivery and Service Discovery in Networks 39

Page 40: My PhD thesis defense presentation

Related Work

• Gill et all [Gill et al., 2007] measured Youtube traffic at edge of a campus network• Network characterization and caching analysis of video traffic based on the Youtube

video popularity seen at the edge of the network. • Regional popularity similar to theirs. Ours is from the “center” as a provider gives us

greater insight into global traffic patterns across the world. • Our findings apply not only to one edge network, but are pretty similar to those seen

across the world

• [Adhikari et al., 2010] measures YouTube statistics at an ISP• The authors find that YouTube employs a location-agnostic, load-balanced method

for delivering video content. • Analyze the early-exit routing of YouTube traffic in ISP network, and find that

YouTube traffic has what they call a “locality bias” (videos showing popularity mostly within a region.)

Improving Content Delivery and Service Discovery in Networks 40

Page 41: My PhD thesis defense presentation

Conclusion

Improving Content Delivery and Service Discovery in Networks 41

Page 42: My PhD thesis defense presentation

Conclusions

• The Internet envisioned 40 years ago in the 1970s is vastly different from what its being used for today.• Rise of powerful handheld / mobile devices

• Consumption of large volumes of data, due to multimedia content

• Need solutions at three layers: networking, application, library

• My dissertation covered some possible solutions, as well as analysis• Opportunistic networks: 7DS, BonAHA

• Content delivery: On-path CDN, ActiveCDN

• Content centric networks: CCNxServ

• Real world video traffic patterns: performed while at JWPlayer

Improving Content Delivery and Service Discovery in Networks 42

Page 43: My PhD thesis defense presentation

List of peer-reviewed publications• Conferences

• Suman Srinivasan, Amandeep Singh, Dhruva Batni, Jae Woo Lee, Henning Schulzrinne, Volker Hilt, Gerald Kunzmann, "CCNxServ: Dynamic Service Scalability in Information-Centric Networks", IEEE ICC 2012, Ottawa, Canada, June 2012.

• Suman Srinivasan, Ivica Rimac, Volker Hilt, Moritz Steiner, Henning Schulzrinne, "Unveiling the content-centric features of TCP", IEEE ICC 2011, Kyoto, Japan, June 2011.

• Jae Woo Lee, Roberto Francescangeli, Jan Janak, Suman Srinivasan, Salman Abdul Baset, Henning Schulzrinne, Zoran Despotovic, Wolfgang Kellerer, "NetServ: Active Networking 2.0", IEEE ICC 2011, Kyoto, Japan, June 2011.

• Kiwoon Sung, Suman Srinivasan, Henning Schulzrinne, "BBS-ONE: Bulletin Board and Forum System for Mobile Opportunistic Networks", IEEE WCNIS 2010, Beijing, China, June 2010.

• Suman Srinivasan, Jae Woo Lee, Eric Liu, Mike Kester, Henning Schulzrinne, Volker Hilt, Srini Seetharaman, Ashiq Khan, "NetServ: Dynamically Deploying In-network Services", ACM ReArch'09 (CoNEXT workshop), December 2009.

• Se Gi Hong, Suman Srinivasan, Henning Schulzrinne, "Measurements of Multicast Service Discovery in a Campus Wireless Network", IEEE Globecom 2009, December 2009.

• Suman Srinivasan, Arezu Moghadam, Henning Schulzrinne, "BonAHA: Service Discovery Framework for Mobile Ad-Hoc Applications", IEEE Consumer Communications & Networking Conference 2009 (CCNC'09), Las Vegas, USA, January 2009.

• Arezu Moghadam, Suman Srinivasan, Henning Schulzrinne, "7DS - A Modular Platform to Develop Mobile Disruption-tolerant Applications", IEEE NGMAST 2008, Wales, UK, September 2008.

• Se Gi Hong, Suman Srinivasan, Henning Schulzrinne, "Accelerating Service Discovery in Ad-hoc Zero Configuration Networking", IEEE Conference on Global Communications (GLOBECOM), Washington D.C., Nov 2007.

• Suman Srinivasan, Arezu Moghadam, Se Gi Hong, Henning G Schulzrinne, "7DS - Node Cooperation and Information Exchange in Mostly Disconnected Networks", IEEE International Conference on Communications (ICC), Glasgow, Scotland, Jun 2007.

• Posters

• Suman Srinivasan, Henning Schulzrinne, "IPv6 Addresses as Content Names in Information-Centric Networking",USENIX ATC 2011 - Poster session, Portland, OR, Jun 2011.

• Suman Srinivasan, Dhruva Batni, Volker Hilt, Henning Schulzrinne, "Dynamic Service Scalability in Information-Centric Networks", USENIX ATC 2011 - Poster session, Portland, OR, Jun 2011.

• Suman Srinivasan, Henning Schulzrinne, "BonSwing: A GUI Framework for Ad-Hoc Applications Using Service Discovery", ACM CoNEXT Student Workshop, New York, Dec 2007.

• Patents

• Volker Hilt, Ivica Rimac, Suman Srinivasan, Moritz Steiner, "Method For Providing On-Path Content Distribution", January 2010

• Henning Schulzrinne, Suman Srinivasan, "Methods And Media For Exchanging Data Between Nodes Of Disconnected Networks", 2008

Improving Content Delivery and Service Discovery in Networks 43

Page 44: My PhD thesis defense presentation

Backup Slides

Improving Content Delivery and Service Discovery in Networks 44

Page 45: My PhD thesis defense presentation

7DS: Check with Peers

Internet

No Internet connection: Check with peers

Improving Content Delivery and Service Discovery in Networks 45

Page 46: My PhD thesis defense presentation

7DS, BonAHA: Technical details

Columbia?

SRV : query._7ds._udp

TXT : columbia

TXT : news

TXT : new york

Times Square?

SRV : query._7ds._udp

TXT : deals

TXT : times square

TXT : weather

“Here is the result for times square”“Here is the result for

columbia”

Improving Content Delivery and Service Discovery in Networks 46

Page 47: My PhD thesis defense presentation

7DS Architecture

Web user

interface

Mail

Transport

Agent

Bulletin

Board

Multicast

Website

Exchange

Web

server

Proxy

server

CachingLogging Configuration

Data

structures

Service

discovery

File

synchronization

Components

Support

services

APIs

Searching

DatabaseXML Parsing

Improving Content Delivery and Service Discovery in Networks 47

Page 48: My PhD thesis defense presentation

7DS Search Engine

• Provides ability to query self for results

• Searches the cache index using Swish-e library

• Presents results in any of three formats: HTML, XML and plain text

• Similar in concept to Google Desktop

Improving Content Delivery and Service Discovery in Networks 48

Page 49: My PhD thesis defense presentation

BonAHA Applications

Group chat• Allows users to discover peers in

local network and chat• Rooms can be set up for private

chats

File Sharing• Users can share files with each

other by dragging and dropping files onto peers’ computers

• Handles peers entering and leaving network

Improving Content Delivery and Service Discovery in Networks 49

Page 50: My PhD thesis defense presentation

CDN HTTP redirection: Youtube Callflows

Original HTTP

request for video

HTTP 301 redirect

to Google Video

server (CDN)

HTTP 301 redirect

to IP address

Improving Content Delivery and Service Discovery in Networks 50

Page 51: My PhD thesis defense presentation

On-path CDN: TCP SYN packets

ProxyBrowser

End-user computerNetSerV router #1 NetSerV router #n

URL requestModified TCP SYN

Name resolution

Forwarded in direction of origin server

Modified TCP SYN

Has content

Sends TCP

SYN-ACK

TCP SYN-ACK

TCP ACK

TCP traffic

Improving Content Delivery and Service Discovery in Networks 51

Page 52: My PhD thesis defense presentation

On-path CDN: Packet trace using Wireshark

Modified SYN packet sent

out successfully

Successful HTTP connection set up

Improving Content Delivery and Service Discovery in Networks 52

Page 53: My PhD thesis defense presentation

On-path CDN: Performance Analysis SetupHow do we know what impact on-path CDNs have?

• Can it reduce delays?

Did a simple RTT analysis

Measurement of round-trip time delays

• CNN server

• Akamai media server

• ALU edge server

• ALU internal node

Traceroute graph generated using Python Scapy’s

traceroute tools

Other ALU nodes

CNN server Akamai server

ALU edge node

ALU in-network node

Improving Content Delivery and Service Discovery in Networks 53

Page 54: My PhD thesis defense presentation

NetServ overview

• Modularization• Building Blocks

• Service Modules

• Virtual services framework• Security, Portability

• NSF FIND four-year project• Columbia University

• Bell Labs

• Deutsche Telekom, DoCoMo Europe

Extensible architecture for core network services

No more ossification in NGIImproving Content Delivery and Service Discovery in Networks 54

Page 55: My PhD thesis defense presentation

ActiveCDN: GENI demo

• NetServ was chosen as one of nine nationwide research projects for NSF GENI testbed

• Demo’d at NSF GENI 2010

• User 1 requests content, gets it directly from server• Meanwhile publisher sends

notification to NetServ node to install module

• User 2 requests content, gets the processed content (with watermark) from NetServ node

Improving Content Delivery and Service Discovery in Networks 55

Page 56: My PhD thesis defense presentation

Content Centric Networking (CCNx)

• Described briefly in related work for CDNs

• Based on Van Jacobson’s (Xerox PARC) work on CCNs• “Can we create a network architecture based

on named data instead of named hosts?”

• Content, instead of URLs• Allows for content mobility, caching, etc.

• Client makes a request for ccnx://content , gets the data without worrying about which node it is on

Improving Content Delivery and Service Discovery in Networks 56

Page 57: My PhD thesis defense presentation

CCNxServ: Architecture

1. Application requests a service on some content

2. CCNxServ converts request into request for content and service module

3. Fetches content and module, and runs the service module to content

4. Puts serviced / processed content back into CCNx space for future requests

Improving Content Delivery and Service Discovery in Networks 57

Page 58: My PhD thesis defense presentation

CCN over IPv6

• CCNx requires a complete modification of underlying network: no more IP

• Rather than modify complete underlying IP network: why not re-use existing IPv6 network?

• IPv6 can hold 2^128 addresses: carve up address space to handle domains, URLs and content requests

• Suman Srinivasan, Henning Schulzrinne, "IPv6 Addresses as Content Names in Information-CentricNetworking", USENIX ATC 2011 - Poster session, Portland, OR, Jun 2011.

Improving Content Delivery and Service Discovery in Networks 58

Page 59: My PhD thesis defense presentation

Related Work

• SCAFFOLD [Freedman et al., 2010]: multiple service instances represented by one common name• Specific service represented through serviceIDs and selected through anycast routing through service routers.

• MILNGENI (million-node GENI) [University of Washington, 2011]: services can be deployed and run on top of many end-systems that are connected via the experimental GENI testbed. • Attempt to deal with the problem of addressing content and making content requests efficient, but operate

on top of the IP layer and require host-to-host communication.

• Service-centric networking (SCN) [Braun et al., 2011]: build a network that runs with services as the primary construct, rather than content. • Superset of CCNx: modification of underlying CCNx implementation.• CCNxServ: builds a service platform on top of only CCNx, implements a fully service-oriented networking

architecture on top of a pure ICN stack.

• SoCCeR [Shanbhag et al., 2011]: build a service layer on top of CCNx. • Control layer to manipulate the underlying routing layer in CCNx, thereby performing distributed best-service

selection using an ant colony optimization approach.

Improving Content Delivery and Service Discovery in Networks 59

Page 60: My PhD thesis defense presentation

• Data movement towards services• Volley [Agarwal et al., 2010]: automate application data placement across data

centers efficiently. • PADS [Belaramani et al., 2009]: a data plane mechanism for transmitting data and

maintaining consistency in large distributed applications and data centers. • Time-shifted TV [Li and Simon, 2011] uses CCNx to improve localized and cooperative

caching using content routers.

• Separation of ID and content name• [Atkinson et al., 2010] and [Dannewitz et al., 2010] aim to preserve privacy of

content names in the network. Almost all of these would require some sort of credential validation to be done on the network.

• LISP (Location/ID Separation Protocol) [Farinacci, 2013] and [IETF, ] is an IETF proposal that deals with a naming system that would configure host-names and machine network address identifiers and map them to locators.

Related Work

Improving Content Delivery and Service Discovery in Networks 60

Page 61: My PhD thesis defense presentation

• Yu et al [Yu et al., 2006] study weekly access patterns of usage and infer arrival rates of video viewers as well as metrics such as session length, popularity and popularity distribution of popular content. • Done at the content provider. Similar to ours for most metrics such as weekly access

patterns, popularity distribution, etc, with the only difference being in total number of data points.

• Their study was done in early 2006, and ours is more current (2012) and at a time when Internet video is more mature.

• Due to video engagement data, in addition to raw video loads, we are able to measure much more interesting trends such as user engagement (such as transition between play, stop, seek and progress states), which gives us more insight into the interplay between the viewer, the provider and the network.

• Costa et al [Costa et al., 2004] study usage patterns of video viewership. • Data is several years old (2004).

Related Work

Improving Content Delivery and Service Discovery in Networks 61

Page 62: My PhD thesis defense presentation

JWPlayer Video: Play by Hour by Country

• Video plays across country of viewing by hour of day

• Traffic patterns are very similar across countries• Video consumption dips

significantly during wee hours of the morning (esp. 4-5 AM)

• Picks up during waking hours, stays steady for most of the day

• Goes up slightly during night time, particularly before / after dinner?

Improving Content Delivery and Service Discovery in Networks 62