Life & Work of Dr. Vinton Cerf and Dr. Robert Kahn | Turing100@Persistent

www.persistentsys.com

Dr. Vinton Cerf and Dr. Robert Kahn
Their Work and its Impact
R. Venkateswaran
CTO, Persistent Systems Ltd
Email: venki@persistent.co.in

© 2012 Persistent Systems Ltd

Agenda

 Bio of Turing Award Recipients – Dr. Vinton Cerf and Dr.
Robert Kahn

 History of the Internet

 Early years of Packet Switching

 Evolution of Networking protocols over the 40 years

 Future architectural options

2

ACKNOWLEDGMENTS:

This presentation has re-used slides and information from
the works of :

• Prof Raj Jain, Washington Univ, St Louis
• Dr. Shivkumar Kalyanaraman, IBM Research, India and
ex-Professor at RPI
• William Stallings – Data & Computer Communication
• Video Interviews of Dr Cerf and Dr Kahn from ACM and
other sources
• Copyrighted material from Robert Zakon’s Internet
Timeline

3

Dr. Vinton Cerf

 Widely known as the “Father of the Internet”
 Illustrious Career
 Elected President of ACM in 2012
 Chief Internet Evangelist, Google
 Chair, ICANN
 Senior VP at MCI, Worldcom
 Worked at DARPA
 Asst Prof at Stanford University

 Awards and Recognition
 ACM Turing Award Recipient – 2004
 Presidential Medal of Freedom in 2005
 Inducted into the Internet Hall of Fame, 2012

 Personal:
 Shares his birthday with Alan Turing
4  Both he and his wife have a hearing impediment © 2012 Persistent Systems Ltd

Dr. Robert Kahn

 Helped design the Packet-Switched ARPANET
 Illustrious Career
 President & CEO of CNRI – Corporation for National Research Initiatives
 Worked at DARPA
 Worked at BBN
 Asst Prof at MIT
 AT&T Bell Laboratories

 Awards and Recognition
 ACM Turing Award Recipient – 2004
 National Medal of Technology, 1997
 Presidential Medal of Freedom, 2005
 Inducted into the Internet Hall of Fame, 2012

 Personal:
 Born on Dec 23, 1938
5

The Internet – 40+ Years

 First 20 Years : Internet 1.0
 Trusted network used by Defense, Research and Academic Communities
 Non-commercial
 Popular applications: Email, FTP, Telnet

 Next 20 Years : Internet 2.0
 Commercial use
 Multiple levels of ownership – leads to distrust and Security concerns
 Wide range of use: Email, WWW, eCommerce, Multimedia
 Key Identifier: IP Address (location, user identity)

 Future : Internet 3.0
 Centered around Users, Data Objects and Hosts
 Service-Oriented
 Builtin Security and Mobility support
6


INTERNET 1.0


Problem addressed by Cerf & Kahn

8

Internetworking Objective

“…both economic and technical
considerations lead us to prefer that the
interface be as simple and reliable as
possible and deal primarily with passing data
between networks using different packet
switching strategies”

V. G. Cerf and R. E. Kahn, 1974

9

Internetworking Architecture

Network Layer
Gateways

10

Foundation for TCP & IP Protocols - I

 Create a nonproprietary universal set of protocols

 Standardized – non-Patented, no constraints or controls

 Universal Addressing across different networks
 Unique identification of hosts in the composite network

 Information slices (packets) are transported from one host to another
via the Internetwork

 Sequence number facilitates ordering of packets

REF: Cerf, V., and R. Kahn, "A Protocol for Packet Network Intercommunication", IEEE Transactions on
Communications, Vol. COM-22, No. 5, pp 637-648, May 1974

11

Foundation for TCP & IP Protocols - II

 Process header (Port) – identify the end process that consumes the
information

 Retransmissions and Duplicate detection – addresses packet loss

 Flow Control – to limit the number of un-acknowledged packets

 Associations between two end processes (“connection”)

 Mechanism to set up & release the association
 3-way handshake, TCP State machine

REF: Cerf, V., and R. Kahn, "A Protocol for Packet Network Intercommunication", IEEE Transactions on
Communications, Vol. COM-22, No. 5, pp 637-648, May 1974

12

ARPANET

13

TCP/IP Layered Architecture Model

End
Application Application
to
end Transport Transport

Network Network Network Network
Hop
by Link Link Link Link Link Link
hop
Physical Physical Physical

Host Router Router Host

14


Network Layer


IP : Protocol of choice for Network Layer

 Packet-switched datagram network
TCP UDP
 IP is the glue (network layer overlay)

 Hourglass architecture IP
 all hosts and routers run IP

Satellite
 Stateless architecture
 no flow state maintained in the Ethernet ATM
network

16

IP Network Design Philosophy

 Keep the network simple
 Simple network devices, intelligent end-systems
 Robust and scalable infrastructure

 IP Networks offer Best-effort service today
 No service guarantees or predictability
 Connectionless, datagram model
 Each IP packet carries sufficient information for its routing
 Hop-by-hop routing – based on destination IP address only
 Packets belonging to the same session could be routed differently
 Out of sequence packet delivery

 Network is application-agnostic – by design

Intelligent end-points, “dumb” network model
17

IP: Minimalist Approach

 Dumb network
 IP provides minimal functionalities to support connectivity
 Addressing, Forwarding, Routing

 Smart end system
 Transport layer or application performs more sophisticated functionalities
 Flow control, Congestion control, Error correction/recovery

 Advantages
 Accommodate heterogeneous technologies
(Ethernet, modem, satellite, wireless)
 Support diverse applications (Email, Telnet, FTP, Usenet)
 Decentralized network administration

18

Packet Switching

Courtesy: William Stallings
19

Circuit Switching

Courtesy: William Stallings
20


Transport Layer


User Datagram Protocol (UDP) – RFC768

 UDP – unreliable, datagram-based protocol

 Connectionless End-to-End Service

 Datagrams may be lost OR delivered out-of-ordered

 Error Detection through Checksum (Optional)

 Minimal overheads – no state maintained

 No Congestion control

 Used by Audio/Video Streams, short-transactional applications

22

Transmission Control Protocol (TCP) –
RFC791

 Designed to provide Reliable process-to-process communication
service in a multi-network environment
 Underlying network unreliable, may deliver bytes out-of-order
 Fragmentation handled at the Network Layer

 Connection-Oriented: Reliable, Ordered, Stream-based, flow and
congestion controlled, bi-directional

 Process Association (State) : combination of IP address and Port
 Maintained only at the end hosts, not in the network
 Sequence number & Window size facilitate ordering and flow control

23

Evolution of TCP

1984
1975 Nagel’s algorithm
Three-way handshake to reduce overhead 1987
Raymond Tomlinson of small packets; Karn’s algorithm 1990
In SIGCOMM 75 predicts congestion to better estimate 4.3BSD Reno
collapse round-trip time fast retransmit
delayed ACK’s
1983
BSD Unix 4.2 1986 1988
1974 supports TCP/IP Congestion Van Jacobson’s
TCP described by collapse algorithms
Vint Cerf and Bob Kahn observed congestion avoidance
In IEEE Trans Comm 1982 and congestion control
TCP & IP (most implemented in
RFC 793 & 791 4.3BSD Tahoe)

1975 1980 1985 1990

Courtesy: Shiv Kalyanaraman
24

TCP Through the 1990s

1994 1996
T/TCP SACK TCP
(Braden) (Floyd et al)
Transaction Selective
TCP Acknowledgement

1993 1994 1996 1996
TCP Vegas ECN TCP Hoe FACK TCP
(Brakmo et al) (Floyd) Improving TCP (Mathis et al)
real congestion Explicit startup extension to SACK
avoidance Congestion
Notification

1993 1994 1996

25

TCP Variants :

 TCP-Tahoe:
 implements the slow start, congestion avoidance, and fast retransmit
algorithms

 TCP-Reno:
 implements the slow start, congestion avoidance, fast retransmit, and fast
recovery algorithms

 Among other implementations are Vegas, NewReno (the most
commonly implemented on webservers today, according to a survey)
and SACK TCP.

26


INTERNET 2.0
Internet goes Commercial


Key Events in early 90s

 Tim-Berners Lee proposed the idea for the World Wide Web
 Mosaic – first graphical web browser launched in 1993

 First commercial dial-up ISP started its service

 Bandwidth doubling every six months

 Push for Multimedia applications

 Move towards Network Unification

28

Some observations

 Demand for broadband access and growth
of bandwidth intensive apps fueling each
other

 Applications demanding more from the
underlying network
 New applications – VoIP, streaming media
 Stringent and often, inflexible requirements
 “One-size fits-all” doesn’t work any more BROADBAND
ACCESS
 “Dumb network” model may have reached its
limits

BANDWIDTH-
INTENSIVE APPS

Networks must provide a more predictable service
29

Growth of the Internet

Copyright: http://www.zakon.org/robert/internet/timeline/
30 Used with Permission © 2012 Persistent Systems Ltd

Growth of the Internet

Copyright: http://www.zakon.org/robert/internet/timeline/
31 Used with Permission © 2012 Persistent Systems Ltd

Applications: network requirements

High
Streaming Interactive
Video Video Conferencing

E-mail with
Attachments
Voice
Requirements

Internet/
Bandwidth

intranet
E-commerce
Text ERP
e-mail
Terminal Mode
Transactions
Low
Low Latency Sensitivity High

32

Evolution of Multi Service Networks

 As Internet Usage increased, an evolved Next Generation Network
(NGN) architecture became important

 Characteristics of the NGN
 One common network capable of handling data, voice and video
communications
 Packet-Switched Network -- “Data friendly” transport and switching
infrastructure
 Flexible services control elements to enable voice communications and
support data and QoS in the future
 Voice parity with the PSTN in terms of features and quality

Source: http://tmdenton.com/pub/bellheads.pdf
33


BellHeads vs NetHeads


Ideological Differences

BellHeads NetHeads
Circuit-switched Telco background IP-based Internet background

Align with Formal ITU Standards Align with Open IETF Standards
Closed Communities Open and Free Community

Believe in Guaranteed QoS Believe in Best-Effort Services

Expect to get Paid for delivering service Expect services to be Cheap/Free

Voice: Per-minute charging with Voice: Just another App on the Packet
settlement between carriers switched network and is FREE
Dumb end-points, Smart Networks Dumb networks, Smart end-points
Wish to Control the Future of the Watch the Network Evolve organically
Network
Prefer Strong Regulatory Environment Open and Unregulated Playing Field

35


Asynchronous Transfer Mode (ATM)
Networks

BellHead view of NGN


Bellheads view of NGN – ATM Networks

 Packet switched network, but connection-oriented

 Fixed size (53-byte) packets called “CELLS”

 Path setup from source to destination before data transfer
(“connection”)

 All Cells of a connection traverse the same path

 In the network, cells are “switched” based on the fixed header (VPI/VCI)

 Using Admission Control policies, network can guarantee QoS for
various applications

37  Could support extremely high speeds (OC-192 or ~10Gbps)

ATM Everywhere

38


Challenge for the Netheads


Per-packet processing in an IP Router

1. Accept packet arriving on an incoming link.

2. Lookup packet destination address in the forwarding table, to
identify outgoing port(s) – Longest Prefix Match

3. Manipulate packet header: e.g., decrement TTL, update
header checksum.

4. Send (switch) packet to the outgoing port(s).

5. Classify and buffer packet in the queue.

6. Transmit packet onto outgoing link.

40

Lookup Rates Required

Optical Line Rate 40 Byte Pkts Lookup
Interface (Gbps) (Mpps) Rates (ns)
OC-12 0.622 1.94 515

OC-48 2.5 7.81 128

OC-192 10.0 31.25 33

OC-768 40.0 125 8

41

Routing Table Size

Source: Geoff Huston, Internet Protocol Journal, Vol 4, No. 1
42

Router Performance

Source: Ipsilon Networks
43


Break-through for the Netheads 


Routing vs Switching

 Routing – based on address lookup and longest prefix match
 Search and compare operation
 Complexity : O(log2 N)

 Switching – based on a circuit number (pre-set)
 Indexing operation
 Complexity : O(1)
 Extremely well-suited for High-speed Networks with Large Address Spaces

45

1996: Ipsilon’s IP Switching Concept

Hybrid: IP routing (control plane) + ATM switching (data plane)
46

Ipsilon’s Solution

47

Ipsilon’s IP Switching

48

TCP/UDP Flow Statistics

49

MPLS: Best of Both Worlds

PACKET CIRCUIT
HYBRID SWITCHING
ROUTING

IP MPLS ATM TDM
+IP

50


Basics of MultiProtocol Label
Switching (MPLS)


MPLS – Multiprotocol Label Switching

 Introduces a new fixed-length header (label) for the IP packet payload
 MPLS is considered Layer 2.5
 Separates forwarding information (label) from IP header
 Easy to implement this in hardware

 Routing at the edge, switching at the core
 MPLS label introduced at the ingress edge
 Core routers use these labels to switch packets
 Label removed at the egress edge router

 Pre-defined paths for switching MPLS packets
 Offline path creation – eliminates path computation overheads

 MPLS header supports CoS markings
52

MPLS Header

Label (20-bits) CoS S TTL

L2 Header MPLS Header IP Packet

32-bits

 Fields
 Label
 Experimental (CoS)
 Stacking bit
 Time to live
 IP packet is encapsulated at an Entry into MPLS domain
 IP packet is de-encapsulated at exit from MPLS domain
53

MPLS – Main Ideas

 Separate forwarding information (label) from the content of IP header

 Single forwarding paradigm (label swapping) - multiple routing
paradigms

 Multiple link-specific realizations of the label swapping forwarding
paradigm

 Flexibility of forming Forwarding Equivalence Classes (FECs)

 Forwarding hierarchy via label stacking

54

MPLS Terminology

Connection Table
In Out Label
IP 25 (port, label) (port, label) Operation
Port 1 Port 2
(1, 22) (2, 17) Swap
(1, 24) (3, 17) Swap
IP 19 (1, 25) (4, 19) Swap
Port 3 Port 4
(2, 23) (3, 12) Swap

 Label Swapping
 Connection table maintains mappings
 Exact match lookup
 Input (port, label) determines:
 Label operation
 Output (port, label)
 Same forwarding algorithm used in Frame Relay and ATM
Courtesy: Rahul Aggarwal
55

MPLS Terminology

Egress
LSR
Ingress
New York
LSR Transit
San LSR Transit
Francisco LSR

LSP
 Ingress LSR (“head-end LSR”)
 Examines inbound IP packets and assigns them to an FEC
 Generates MPLS header and assigns initial label
 Transit LSR
 Forwards MPLS packets using label swapping
 Egress LSR (“tail-end LSR”)
56
 Removes the MPLS header Courtesy: Rahul Aggarwal

MPLS Forwarding Model

Source Egress
Ingress LSR Paris
LSR

Rome

 Ingress LSR determines FEC and assigns a label
 Forwards Paris traffic on the Green LSP
 Forwards Rome traffic on the Blue LSP
 Traffic is label swapped at each transit LSR
 Egress LSR
 Removes MPLS header
 Forwards packet based on destination address Courtesy: Rahul Aggarwal
57

MPLS Forwarding vs. IP Routing

Source Destination
IP Routing Domain

Examine IP header Examine IP header Examine IP header Examine IP header
Assign to FEC Assign to FEC Assign to FEC Assign to FEC
Forward Forward Forward Forward

Ingress Egress
Source Destination
LSR MPLS Domain LSR

Examine IP header Examine IP header
Assign to FEC Label swap Label swap Assign to FEC
Forward Forward Forward Forward
58

MPLS Forwarding Example

MPLS Table
In Out 134.5.6.1
(2, 84) (6, 0)

134.5.1.5
2 6
Egress Routing Table
Destination Next Hop
200.3.2.7 2
134.5/16 134.5.6.1

3 200.3.2/24 200.3.2.1

Ingress Routing Table 1 2 3 5
Destination Next Hop 200.3.2.7
134.5/16 (2, 84)

200.3.2/24 (3, 99)

MPLS Table MPLS Table 200.3.2.1 200.3.2.7
In Out In Out
(1, 99) (2, 56) (3, 56) (5, 0)

59

MPLS - Summary

 MPLS forwarding algorithm is simpler than IP forwarding algorithm
 Fixed size header vs variable sized header

 Enables more functionality than could be provided with the IP
forwarding algorithm
 Eg. All packets traverse the same path through the domain

 MPLS is an architectural shift
 Enables many applications; difficult to foresee all of them

 Most Service Providers today are moving towards MPLS deployment
 Reduce cost, simplify operations, introduce high-value services

 MPLS is also evolving
 ER-MPLS, MPLS-TE etc.

60


INTERNET 3.0
End of Network-Centric viewpoint


Some challenges with today’s Internet

 Support for Mobility with TCP/IP not efficient

 Security issues – viruses, spams, bots, DDOS attacks, hacks
 Internet was designed with “friendly” users in mind

 Multi-homing not well supported by TCP/IP
 Change in IP Interface results in service disruption

 Applications are designed to work well *only* if round-trip delays are
small

 TCP/IP expects most parts of the network to be interconnected

 Routing protocols are becoming complex
Courtesy: Raj Jain
62  Routers are becoming expensive

Challenges….

 Network-centric (“Where”) approach not optimal for various applications
 People-centric – “Who”
 Content-centric – “What”

 IP Address correlates Identity and Location
 This may neither be necessary nor desirable

Courtesy: Raj Jain
63

Future Research

 Delay Tolerant Network Architecture

 Content-Centric Networks

 Software Defined Network (SDN)
 Network Virtualization paradigm

64


Delay Tolerant Network


Delay Tolerant Networks (DTN)

 DTNs Characterized by
 Intermittent Connectivity
 Extremely Long Delays
 Asymmetric Data Rates
 High Error Rates

 Addressed using a Store-and-Forward Message Switching paradigm
 IETF RFC 4838

66

Delay Tolerant Network – Overlay Arch

Source: http://www.cs.rice.edu/~scrosby/TA/comp620-s05/papers/F03.pdf
67


Content Centric Network


Content Centric Networks

69

Content Centric Networks

 Strategy – Figuring out the best path to deliver the content
 Dynamic optimization

 Security – Ensure content access only to authorized users

 Interest Packet – used to request a specific content

 Data Packet – contains the specific content

 Mechanisms to forward an Interest packet towards available content
store

70


Software Defined Network


SDN Architecture

App App App App
Open API
Network Operating System

Open Interface
to Hardware
(OpenFlow)
Openflow Openflow
Firmware Firmware
Packet-Forwarding Packet-Forwarding
Hardware Hardware

Openflow Openflow
Firmware Firmware
Packet-Forwarding Packet-Forwarding
Hardware Hardware
Courtesy: Matt Davy
72

Analogy with Computer Industry

Computer Industry Network Industry

Apps Apps Apps Apps Apps Apps

Network
Windows Linux FreeBSD NOX Beacon
OS

Virtualization Virtualization

Openflow
x86

Courtesy: Matt Davy
73

Software Defined Network

74

Does this look familiar?

75


THANK YOU
Email: venki@persistent.co.in


References

1. Prof Raj Jain’s home page - http://www.cse.wustl.edu/~jain/
2. Dr Shivkumar Kalyanaraman’s lectures -
http://www.ecse.rpi.edu/Homepages/koushik/shivkuma-
teaching/video_index.html
3. Hobbes’ Internet Timeline - http://www.zakon.org/robert/internet/timeline/
4. ACM Turing Award Video Lecture of Dr Cerf & Dr Kahn -
http://amturing.acm.org/vp/cerf_1083211.cfm
5. Vint Cerf & Bob Kahn’s seminal paper on Protocol for Packet Network
Intercommunication
http://www.cs.princeton.edu/courses/archive/fall06/cos561/papers/cerf74.pdf
6. Relevant RFCs from IETF - www.ietf.org

77

Life & Work of Dr. Vinton Cerf and Dr. Robert Kahn | Turing100@Persistent

Recomendados

Recomendados

Más contenido relacionado

La actualidad más candente

La actualidad más candente (18)

Similar a Life & Work of Dr. Vinton Cerf and Dr. Robert Kahn | Turing100@Persistent

Similar a Life & Work of Dr. Vinton Cerf and Dr. Robert Kahn | Turing100@Persistent (20)

Más de Persistent Systems Ltd.

Más de Persistent Systems Ltd. (20)

Último

Último (20)

Life & Work of Dr. Vinton Cerf and Dr. Robert Kahn | Turing100@Persistent