The forces behind the changing Internet: IXPs, content delivery …steve/presentations/Content-delivery-COMM… · Diversity & traffic engineering • Diversity in content location

Steve Uhlig 1

The forces behind the changing Internet: IXPs, content delivery and virtualization

Steve Uhlig Professor of Networks

Head of Networks research group Queen Mary, University of London

[email protected] http://www.eecs.qmul.ac.uk/~steve/

Credit to collaborators: Bernhard Ager, Nikos Chatzis, Anja Feldmann, Benjamin Frank, Bruce

Maggs, Wolfgang Mühlbauer, Ingmar Poese, Nadi Sarrar, Georgios Smaragdakis, Walter Willinger

Steve Uhlig 2

Internet Infrastructure

Internet: communication platform

Internet: content delivery platform

Steve Uhlig 3

Agenda

•  Internet update •  Short history of content delivery •  Content delivery 101 •  Network virtualization •  Future Internet

Steve Uhlig 4

The Beast

4

Steve Uhlig 5

Accepted view of the Internet

•  35,000+ networks •  Hierarchical structure -  Tier-1 (10-20): ATT, L3, Sprint,… -  Regional ISPs (15%): BT, Telefonica,… -  Stubs (85%): eyeball ISPs, universities,

enterprise networks

•  Known AS connectivity -  Customer-provider: 90,000+ -  Peer-peer: 35-40,000

Steve Uhlig 6

Old mental model

Steve Uhlig 7

Most recent mental model

C. Labovitz, S. Iekel-Johnson, D. McPherson, J. Oberheide, and F. Jahanian. Internet Interdomain Traffic. Proc. of ACM SIGCOMM 2010.

Steve Uhlig 8

Internet Exchange Point •  An Internet exchange point (IXP)

is a layer 2 service to facilitate the interconnection between -  ISPs -  Hosting or service providers -  CDNs

•  An IXP facilitates peering between players, usually across a public and/or private peering fabric of some type

•  Offer public and/or private peerings

William B. Norton. The Internet Peering Playbook : Connecting to the Core of the Internet. DrPeering Press, 2012.

Steve Uhlig 9

IXP architecture: example

AS6AS7

AS8

AS9... ASN

AS1 AS2 AS3AS4

AS5

IXP switches

Members' routers

B. Ager, N. Chatzis, A. Feldmann, N. Sarrar, S. Uhlig, and W. Willinger. Anatomy of a Large European IXP. Proc. of ACM SIGCOMM 2012.

Steve Uhlig 10

Interconnection and business

William B. Norton. The Internet Peering Playbook : Connecting to the Core of the Internet. DrPeering Press, 2012.

Steve Uhlig 11

Peerings at an IXP

0 50 100 150 200 250 300 350 400

050

150

250

350

Members

Num

ber o

f pee

rs ●

●

●

●

●●

●●

●

●

●●

●

●●

●

●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●●●●

●

●

●

●

●

●

●

●

●

●

●

●

●●●

●

●

●

●

●

●●

●

●

●

●

●

●

●●●●

●

●

●●

●

●

●●

●

●

●

●

●

●

●

●●

●

●

●

●

●●

●

●

●

●

●

●

●●

●

T1 T2 Leaf●LISP SISP HCDN AEN

Steve Uhlig 12

Updated view of the Internet

•  Mixed structure -  Tier-1 (10-20) + Large IXPs -  Regional ISPs + smaller regional IXP

•  Known AS connectivity -  Customer-provider: 90,000+ -  Peer-peer: data from a single IXP doubles it!

•  Traffic -  Increasingly exchanged directly between CDN

and regional ISPs B. Ager, N. Chatzis, A. Feldmann, N. Sarrar, S. Uhlig, and W. Willinger. Anatomy of a Large European IXP. Proc. of ACM SIGCOMM 2012.

Steve Uhlig 13

Content is power

•  CAIDA: BGP-based degree or customer-cone •  Renesys: variant of CAIDA-cone •  Knodes: Fixedorbit.com centrality metric •  Arbor: Interdomain traffic •  Potential: hostname-based •  Normalized potential: weighted hostnames

B. Ager, W. Mühlbauer, G. Smaragdakis, and S. Uhlig. Web content cartography. Proc. of ACM SIGCOMM IMC 2011.

Steve Uhlig 14

Agenda


Steve Uhlig 15

The Early Web

Pathan Mukaddim. Ongoing Trends and Future Directions in Content Delivery Networks (CDNs). Available online from: http://amkpathan.wordpress.com/article/ongoing-trends-and-future-directions-in-3uxfz2buz8z1w-2/

A

Steve Uhlig 16

CDNs 1.0


A

Steve Uhlig 17

CDNs 2.0


A

Steve Uhlig 18

Autonomic CDNs


Steve Uhlig 19

Agenda


Steve Uhlig 20

Foundations

•  Server selection -  HTTP (DNS)

-  P2P

•  Deployment -  HTTP: data centers and CDNs

-  P2P: swarms

•  Exposed network location diversity

Steve Uhlig 21

HTTP server selection

DNS

HTTP

Steve Uhlig 22

DNS responsiveness

10 20 50 100 500 2000 5000

1e−0

41e−0

21e

+00

Response time (ms)

CCDF

GOOGLE 1stLOCAL 1stOPENDNS 1stGOOGLE 2ndLOCAL 2ndOPENDNS 2nd

Good ISP

10 20 50 100 500 2000 5000

1e−0

41e−0

21e

+00

Response time (ms)

CCDF

GOOGLE 1stLOCAL 1stOPENDNS 1stGOOGLE 2ndLOCAL 2ndOPENDNS 2nd

Bad ISP

B. Ager, W. Mühlbauer, G. Smaragdakis, and S. Uhlig. Comparing DNS resolvers in the wild. Proc. of ACM SIGCOMM IMC 2010.

Steve Uhlig 23

DNS caching behavior

10 20 50 100 500 2000 5000

1020

5010

020

050

020

0050

00

Time for first query (ms)

Tim

e fo

r sec

ond

quer

y (m

s)

GOOGLELOCALOPENDNS

Good ISP

10 20 50 100 500 2000 5000

1020

5010

050

020

0050

00

Time for first query (ms)

Tim

e fo

r sec

ond

quer

y (m

s)

GOOGLELOCALOPENDNS

Bad ISP

Steve Uhlig 24

3rd party DNS resolvers

020

040

060

080

0

vantage points(sorted by # returned IPs that are in same AS)

# re

turn

ed IP

s th

at a

re in

sam

e AS

x

GoogleLocalOpenDNS

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Steve Uhlig 25

DNS deployment

•  ISP DNS deployment: -  Fixed: few bad ISPs -  Mobile: many bad ISPs

•  3rd party DNS: -  Raw performance: good compared to bad ISPs -  Confusing to content delivery networks!

Steve Uhlig 26

Foundations


-  P2P


-  P2P: swarms


Steve Uhlig 27

Server selection: P2P Tracker

Client

Swarm

Steve Uhlig 28

Server selection: P2P

•  Today: -  More or less random, e.g., Bittorrent unchokes 4

best peers based on average download rate •  Alternatives: -  PADIS: provider-aided distance information

service -  P4P: provider-aided P2P selection -  Biased Unchoking: consider locality of peers in

their selection -  Ono: Vuze plugin, use CDN measurements to

guess peer proximity •  Plenty of cool research hiding!

Steve Uhlig 29

Foundations


-  P2P


-  P2P: swarms


Steve Uhlig 30

World data centers

http://www.datacentermap.com/

Steve Uhlig 31

Google data centers

http://royal.pingdom.com/2008/04/11/map-of-all-google-data-center-locations/

Steve Uhlig 32

World clouds

http://www.datacentermap.com/

Steve Uhlig 33

Today’s popular CDNs

B. Ager, W. Mühlbauer, G. Smaragdakis, and S. Uhlig. Web content cartography. Proc. of ACM SIGCOMM IMC 2011.

Steve Uhlig 34

CDNs geographic coverage

Steve Uhlig 35

Where is content?

•  California •  China 2nd! •  USA: 9 among top

20 •  Other developed

countries: low normalized potential

Steve Uhlig 36

Foundations


-  P2P


-  P2P: swarms


Steve Uhlig 37

Deployment of popular P2P swarms

Steve Uhlig 38

Foundations


-  P2P


-  P2P: swarms


Steve Uhlig 39

CDN diversity

0

0.2

0.4

0.6

0.8

1

1 10 100

CD

F of

tota

l HTT

P Tr

affic

Available Subnets

Steve Uhlig 40

P2P diversity

Steve Uhlig 41

Diversity & traffic engineering

•  Diversity in content location is an opportunity for traffic engineering

•  E.g., content-aware traffic engineering

B. Frank, I. Poese, G. Smaragdakis, S. Uhlig, and A. Feldmann. Content-aware traffic engineering. Proc. of ACM SIGMETRICS 2012.

ISPCP Server A CP Server B

CP Server C

ClientClientClientClient

Highly utilizedlink

shift

Steve Uhlig 42

Agenda


Steve Uhlig 43 43

The Virtualized Network

Virtualization Management

Provisioning of Virtual Networks (on - demand instantiation of virtual networks)

Infrastructure

Virtualized Substrate

Virtual Network Virtual

Network

Virtualization of Resources (partitioning of physical infrastructure into “ slices ” )

Virtualization Management

Provisioning of Virtual Networks (on - demand instantiation of virtual networks)

Infrastructure

Virtualized Substrate

Virtual Network Virtual

Network

Virtualization of Resources (partitioning of physical infrastructure into “ slices ” )

Steve Uhlig 44 44

Why virtualization?

Virtual network = resource isolation/sharing •  Different architecture/protocol per virtual

network -  Does not have to be IP protocol, e.g., CCN -  For QoS, security, different types of content/

applications •  Expose network components to applications

and services •  Dynamic: migration/expansion/contraction

Steve Uhlig 45 45

Benefits

•  Isolation as enabler for new technologies -  Away from traditional native services, e.g. IPv6,

multicast -  Deployment of innovative services

•  Agile network management -  Migration of devices (such as routers) -  Managing a network “as a cloud”

Steve Uhlig 46

•  Beyond today’s monolithic network equipment

•  Separation of control and data plane through software modularity, e.g., Linux

•  Do not change existing control plane Control plane

Forwarding engine

Software

Hardware

Remote controller Communication

channel

•  Principles

-  Communication channel between forwarding engine and remote controller

-  Expose network equipment capabilities, e.g., TCAM, QoS

Software-defined networking

Steve Uhlig 47

Google and SDN

•  Google is using OpenFlow •  Purpose: traffic engineering

Steve Uhlig 48

Agenda


Steve Uhlig 49

CDN 3.0

•  Hybrid infrastructures: Akamai, PPTV •  Meta-CDNs, e.g., Conviva •  Virtual CDNs through ISP micro-datacenters

ISPS

S

S

S

S

S

S

S

= PoP with Microdatacenter (Service deployed)= PoP with Microdatacenter

PoPs

Steve Uhlig 50

Content-Processing-Network Infrastructure

Storage

Processing

Content-Processing-Network

+

+

Steve Uhlig 51

Challenges & open problems

•  Changing Internet ecosystem: measurements •  Massive content infrastructure deployment:

scalability •  Significant location diversity: traffic

engineering •  Increased diversity and agility through

virtualization: built-in adaption

Steve Uhlig 52

References

•  R. Buyya, M. Pathan, A. Vakali (Eds.). Content Delivery Networks. Springer-Verlag Lecture Notes in Electrical Engineering, 2008.

•  H. Wang, J. Liu, and K. Xu. On the locality of BitTorrent-based video file swarming. Proc. of IPTPS 2009.

•  C. Labovitz, S. Iekel-Johnson, D. McPherson, J. Oberheide, and F. Jahanian. Internet Interdomain Traffic. Proc. of ACM SIGCOMM 2010.

•  I. Poese, B. Frank, B. Ager, G. Smaragdakis, and A. Feldmann. Improving content delivery using provider-aided distance information. Proc. of ACM SIGCOMM IMC 2010.

•  B. Ager, W. Mühlbauer, G. Smaragdakis, and S. Uhlig. Comparing DNS resolvers in the wild. Proc. of ACM SIGCOMM IMC 2010.

•  B. Ager, W. Mühlbauer, G. Smaragdakis, and S. Uhlig. Web content cartography. Proc. of ACM SIGCOMM IMC 2011.

Steve Uhlig 53

References

•  William B. Norton. The Internet Peering Playbook : Connecting to the Core of the Internet. DrPeering Press, 2012.

•  P. Dhungel, K. Ross, M. Steiner, Y. Tian, X. Hei. Xunlei: Peer-Assisted Download Acceleration on a Massive Scale. Proc. of PAM 2012.

•  I. Poese, B. Frank, B. Ager, G. Smaragdakis, S. Uhlig, A. Feldmann. Improving Content Delivery with PaDIS. IEEE Internet Computing 16(3): 46-52, 2012.

•  B. Frank, I. Poese, G. Smaragdakis, S. Uhlig and A. Feldmann. Content-aware traffic engineering. Proc. of ACM SIGMETRICS 2012.

•  B. Ager, N. Chatzis, A. Feldmann, N. Sarrar, S. Uhlig, and W. Willinger. Anatomy of a Large European IXP. Proc. of ACM SIGCOMM 2012.

Documents

The forces behind the changing Internet: IXPs, content delivery …steve/presentations/Content-delivery-COMM… · Diversity & traffic engineering • Diversity in content location