PARADOXES IN INTERNET S. Keshav University of Waterloo - - PowerPoint PPT Presentation

▶
paradoxes in internet
SMART_READER_LITE
LIVE PREVIEW

PARADOXES IN INTERNET S. Keshav University of Waterloo - - PowerPoint PPT Presentation

Apr 01, 2024 252 likes •1.21k views

PARADOXES IN INTERNET S. Keshav University of Waterloo ARCHITECTURE Chair, ACM SIGCOMM UNIVERSITY OF WATERLOO UNIVERSITY OF WATERLOO Founded 1957 35,000 students Faculty of Mathematics 250 faculty 8000 undergrads 1000

slide-1
SLIDE 1

PARADOXES IN INTERNET ARCHITECTURE

  • S. Keshav

University of Waterloo Chair, ACM SIGCOMM

slide-2
SLIDE 2

UNIVERSITY OF WATERLOO

UNIVERSITY OF WATERLOO Founded 1957 35,000 students Faculty of Mathematics

  • 250 faculty
  • 8000 undergrads
  • 1000 grads
slide-3
SLIDE 3
slide-4
SLIDE 4

http://www.computerhistory.org/internethistory

slide-5
SLIDE 5

http://www.computerhistory.org/internethistory

slide-6
SLIDE 6
slide-7
SLIDE 7
slide-8
SLIDE 8
slide-9
SLIDE 9
slide-10
SLIDE 10

1969 vs. 2014

slide-11
SLIDE 11

http://www.computerhistory.org/internethistory

1969 vs. 2014

slide-12
SLIDE 12
slide-13
SLIDE 13

http://www.ict-mplane.eu/public/about-mplane-intelligent-measurement-plane-future-network-and-application

slide-14
SLIDE 14

THE VISION FOR COMPUTER NETWORKING

Anytime access to any information by anyone anywhere

slide-15
SLIDE 15

ARE WE DONE?

slide-16
SLIDE 16

MAYBE NOT…

slide-17
SLIDE 17

Images: http://www.cupcakeproject.com/2012/06/homemade

slide-18
SLIDE 18

Images: http://www.cupcakeproject.com/2012/06/homemade-spam

slide-19
SLIDE 19

Images: http://www.cupcakeproject.com/2012/06/homemade-spam-recipe.html

slide-20
SLIDE 20

WHY CAN’T WE SIMPLY BLOCK SPAMMERS?

slide-21
SLIDE 21

BACK TO BASICS…

http://bit.kuas.edu.tw/~csshieh/teach/

slide-22
SLIDE 22

CLIENT SERVER MODEL

You Spammer

slide-23
SLIDE 23

REALITY

You Spammer …

slide-24
SLIDE 24

ISP RELATIONSHIP

You Spammer … Your ISP Spammer’s ISP

slide-25
SLIDE 25

INFORMATION HIDING

You Spammer Your ISP … Spammer’s ISP

slide-26
SLIDE 26

THE REAL PROBLEM

Narrow AS-AS relationship

 Data plane: Packet exchange  Control plane: Route information exchange Identities (and QoS) do not traverse AS boundaries AS behaviour is unregulated beyond packet transfer

slide-27
SLIDE 27
slide-28
SLIDE 28

THESIS

Many of the key problems in the Internet today are due to its

  • rigins as an academic research project

The very things that led to its success lie at the heart of its failures

slide-29
SLIDE 29

BACK TO THE BEGINNING…

Clark, David. "The design philosophy of the DARPA Internet protocols." ACM SIGCOMM Computer Communication Review 18.4 (1988): 106-114

slide-30
SLIDE 30

ORIGINAL DESIGN GOALS

slide-31
SLIDE 31

ORIGINAL DESIGN GOALS

slide-32
SLIDE 32

VERY SUCCESSFUL!

http://fortune.com/2014/06/23/telecom-companies-count-386-billion-in-lost-revenue-to-skype-whatsapp

slide-33
SLIDE 33

HOW TO REDUCE COST?

FACT: Computer communication is inherently bursty CONSEQUENCE: Allocating a circuit (‘phone call’) to it is expensive Cheaper to share (‘multiplex’) a circuit among many end-to-end communications

slide-34
SLIDE 34

A B

slide-35
SLIDE 35

A B C D

slide-36
SLIDE 36

A B C D

slide-37
SLIDE 37

A B C D

slide-38
SLIDE 38

A B C D

slide-39
SLIDE 39

A B C D

slide-40
SLIDE 40

A B C D

slide-41
SLIDE 41

A B C D Drop

slide-42
SLIDE 42

A B C D Enqueue

slide-43
SLIDE 43

A B C D Enqueue But this adds delay!

slide-44
SLIDE 44

A B C D Enqueue Amount of delay depends on the load…

slide-45
SLIDE 45

M/M/1 QUEUEING DELAY

Load Queueing delay

slide-46
SLIDE 46

QUALITY OF SERVICE

Four well-known approaches

Overprovisioning Admission control Differential service quality: prioritize delay-sensitive flows Drop packets when the queue size grows, expecting sources to respond

slide-47
SLIDE 47

QUALITY OF SERVICE

All approaches have serious problems

 Overprovisioning  Expensive  Admission control  Requires end-to-end adoption  Impossible to allocate costs (more later)  Differential service quality: prioritize delay-sensitive flows  Requires changes to scheduling disciplines at every multiplexor  Drop packets when the queue size grows, expecting sources to respond  Requires complex tuning  Assumes cooperation

slide-48
SLIDE 48

BOTTOM LINE

The primary design goal of the Internet makes it inherently unsuitable for real-time communication

slide-49
SLIDE 49

ORIGINAL DESIGN GOALS

slide-50
SLIDE 50

ORIGINAL DESIGN GOALS

Clark, David. "The design philosophy of the DARPA Internet protocols." ACM SIGCOMM Computer Communication Review 18.4 (1988): 106

slide-51
SLIDE 51

THE INTERNET IS A NETWORK OF NETWORKS

slide-52
SLIDE 52

ACCOMMODATING HETEROGENEITY

slide-53
SLIDE 53
slide-54
SLIDE 54

NARROW INTERFACE

slide-55
SLIDE 55

NARROW INTERFACE

Allows interoperability across heterogeneous technologies Easy to implement Allows independent evolution

slide-56
SLIDE 56

VERY SUCCESSFUL

The architecture has survived the transition of individual ASs from dialup lines to multi-lambda optical fibers from text-based interaction to multimedia on wireless devices while retaining interoperability!

slide-57
SLIDE 57

BUT…

Allows interoperability across heterogeneous technologies Easy to implement Allows independent evolution

No support for quality of service

slide-58
SLIDE 58

AND…

Allows interoperability across heterogeneous technologies Easy to implement Allows independent evolution No support for quality of service

Unconstrained implementation Arbitrary layering Impossible to debug performance

slide-59
SLIDE 59

Source: Designing Multi-layer Carrier Networks for Capacity and Survivability, OPNETWORK 2012

slide-60
SLIDE 60

ORIGINAL DESIGN GOALS

slide-61
SLIDE 61

ORIGINAL DESIGN GOALS

Clark, David. "The design philosophy of the DARPA Internet protocols." ACM SIGCOMM Computer Communication Review 18.4 (1988): 106

slide-62
SLIDE 62

ORIGINAL DESIGN GOALS

Clark, David. "The design philosophy of the DARPA Internet protocols." ACM SIGCOMM Computer Communication Review 18.4 (1988): 106

slide-63
SLIDE 63

ORIGINAL DESIGN GOALS

Clark, David. "The design philosophy of the DARPA Internet protocols." ACM SIGCOMM Computer Communication Review 18.4 (1988): 106

slide-64
SLIDE 64

SUPPORTING MULTIPLE SERVICE TYPES

TCP and UDP support a huge variety of protocols An unqualified success! But…

slide-65
SLIDE 65

SUPPORTING MULTIPLE SERVICE TYPES

Even the 1988 paper abandons real-time services

slide-66
SLIDE 66

ORIGINAL DESIGN GOALS

Clark, David. "The design philosophy of the DARPA Internet protocols." ACM SIGCOMM Computer Communication Review 18.4 (1988): 106

slide-67
SLIDE 67

ORIGINAL DESIGN GOALS

slide-68
SLIDE 68

DISTRIBUTED MANAGEMENT

Distributes the task of management using Autonomous Systems

slide-69
SLIDE 69

WEAK CENTRALIZATION

ICANN IANA Registries DNS TLDs

slide-70
SLIDE 70

DISTRIBUTED MANAGEMENT

Allows rapid deployment Allows independent evolution Delegation allows massive scaling

 DNS

slide-71
SLIDE 71

DISTRIBUTED MANAGEMENT

Allows rapid deployment Allows independent evolution Delegation allows massive scaling

 DNS

With narrow interfaces, makes quality of service even more challenging

slide-72
SLIDE 72

DISTRIBUTED MANAGEMENT

Allows rapid deployment Allows independent evolution Delegation allows massive scaling

 DNS

With narrow interfaces, makes quality

  • f service even more challenging

No network-wide identity

Security nightmare Spam, DDOS, hacking, …

slide-73
SLIDE 73

DISTRIBUTED MANAGEMENT

Allows rapid deployment Allows independent evolution Delegation allows massive scaling

 DNS

With narrow interfaces, makes quality

  • f service even more challenging

No network-wide identity

 Security nightmare  Spam, DDOS, hacking, …

No single view into the network

 Makes networks unmanageable

slide-74
SLIDE 74

DISTRIBUTED MANAGEMENT

Allows rapid deployment Allows independent evolution Delegation allows massive scaling

 DNS

With narrow interfaces, makes quality of service even more challenging No network-wide identity

 Security nightmare  Spam, DDOS, hacking, …

No single view into the network

 Makes networks unmanageable

Autonomous systems

 Can inspect, modify, and drop packets  No privacy

slide-75
SLIDE 75

ORIGINAL DESIGN GOALS

slide-76
SLIDE 76

REDUCING ATTACHMENT EFFORT

What is needed to get an endpoint on the telephone network?

Verinymous identity! Endpoint identifier and end-user identity are closely bound

 Allows billing and tracing

slide-77
SLIDE 77

REDUCING ATTACHMENT EFFORT

What is needed to get an endpoint on the Internet ?

IP address, netmask, and IP address of closest router Makes it very easy to attach a node to the Internet

But endpoint identifier and human’s identity are unbound

Enables spam

slide-78
SLIDE 78

ORIGINAL DESIGN GOALS

slide-79
SLIDE 79

ORIGINAL DESIGN GOALS

slide-80
SLIDE 80
slide-81
SLIDE 81

ORIGINAL DESIGN GOALS

slide-82
SLIDE 82

Images: http://www.cupcakeproject.com/2012/06/homemade-spam-recipe.html

slide-83
SLIDE 83
slide-84
SLIDE 84

WHAT TO DO?

slide-85
SLIDE 85
slide-86
SLIDE 86

LET’S REVISIT ONE OF THE GOALS

slide-87
SLIDE 87
slide-88
SLIDE 88

THIS DESIGN APPROACH IS LONG DEAD…

SDN MPLS for traffic shaping Middleboxes

 Load balancers  Firewalls  Intrusion detectors  VPN endpoints  …

slide-89
SLIDE 89

TELEPHONE NETWORK

Can we integrate the best aspects of the Internet with the best aspects of the telephone network?

 Prevent spam by allowing identities to be traced  Require privacy from carriers  Make the inter-AS interface richer to allow QoS

slide-90
SLIDE 90

Dock

Keshav, S.. "Why cell phones will dominate the future internet." ACM SIGCOMM Computer Communication Review 35.2 (2005): 83

slide-91
SLIDE 91

TIME TO RETHINK INTERNET ARCHITECTURE

slide-92
SLIDE 92

TIME TO BE CREATIVE!

Technology trends and future demands

 Industrial Internet of Things  Extreme sensing  In-body Internet  Deep Space Internet  Hackers  Need for privacy  Quality of Service

slide-93
SLIDE 93

TIME TO BE CREATIVE!

Technology trends and future demands

 Industrial Internet of Things  Extreme sensing  In-body Internet  Deep Space Internet  Hackers  Spam  Privacy  Quality of Service

What should be our new design philosophy? How can we design our future networks to be legacy compatible?

slide-94
SLIDE 94