1

Cha

INTERNET ARCHITECTURE

1
Slides are an edited mashup of two books

Computer Networks: Computer

A Systems Approach, 5e Top Down A
Networking:
6t editio
Larry L. Peterson and Bruce S. Approach
h
Jimn Kurose, Keith
Davie Addiso
Ross -Wesle
March
n y
Copyright © 2010, Elsevier Inc. All rights
2012 2
 Reserved J.F Kurose and K.W. Ross, All Rights Reserved
Cha
Why is Internet Architecture relevant to you
n The Internet is our economies most critical infrastructure. n The
Internet is our economies most enabling metaphor.
n Objectives for student learning:
n What are the technical features that allowed the Internet to
become so pervasive and diverse during the course of your
lifetime.
n Understand its basic structure, and distill from it lessons for (a) the
use and governance of Internet technology in modern
organizations, and
(b) lessons for creation of other new technologies and
products.

3
Internet history Cha

1972-1980: Internetworking, new and proprietary

nets
n 1970: ALOHAnet satellite network
in Hawaii Cerf and Kahn’s internetworking
principles:
n 1974: Cerf and Kahn - architecture
n minimalism, autonomy
for interconnecting networks
no internal changes
n 1976: Ethernet at Xerox PARC n required to interconnect
late70’s: proprietary architectures: networks
DECnet, SNA, XNA n late 70’s: n best effort service model

switching fixed length packets (ATM n stateless routers n

precursor) n 1979: ARPAnet has 200 decentralized control
nodes define today’s Internet
architecture

Introduction 4
Internet history Cha

1-4

1980-1990: new protocols, a proliferation of

networks
n 1983: deployment of n 1988: TCP congestion
TCP/IP n 1982: control
smtp e-mail protocol n new national networks:
defined Csnet, BITnet, NSFnet,
n 1983: DNS defined for Minitel n 100,000
name-to-IP-address hosts connected to
translation confederation of
n 1985: ftp protocol networks
defined

Introduction 5
Internet history
1-5

Chapter 1
1990, 2000’s: commercialization, the Web,
new apps
n early 1990’s: ARPAnet n hypertext [Bush 1945, Nelson
decommissioned 1960’s] n HTML, HTTP:
Berners-Lee n 1994: Mosaic, later
n 1991: NSF lifts restrictions
Netscape n late 1990’s:
on
commercialization
commercial use of NSFnet
of the Web
(decommissioned,
1995) n early 1990s: late 1990’s – 2000’s:
Web

Introduction 6
Internet history
n more killer apps: instant Gbps
messaging, P2P file
sharing
n network security to
forefront n est. 50 million
host, 100 million+ users
n backbone links running at 1-6

2005-present n ~750 million

hosts n Smartphones and tablets
n Aggressive deployment of broadband access n Increasing ubiquity of
high-speed wireless access n Emergence of online social networks:
n Facebook, Instagram, …

Introduction 7
Internet history Cha

n Service providers (Google, Amazon) create their own networks n

Bypass Internet, providing “instantaneous” access to search, email,
etc.
n E-commerce, universities, enterprises running their services in
“cloud” (eg, Amazon EC2)

1-7

Introduction 8
 Cha
What makes the Internet appear as single service

n Networks share common architecture and protocols that

enable communication within and among them.
n Architecture: how components of the networks interrelate n
Protocols: standards governing the interchange of data
n The architecture and protocols were and to some extent
are shaped by fundamental (and interrelated) design
principles adopted by early builders of the Internet. n
Hourglass n End-to-End
n Distributed design and decentralized control n
Heterogeneity, Scalability

9
 Chapte
Connectivity Terminologies
n Link, Nodes
n Point-to-point, Multiple access
n Switched Network
n Circuit Switched
(a)
n Packet Switched: Store -
and-forward
n Cloud
n Hosts
n Switches
n internetwork
n Router/gateway
n Host-to-host connectivity
(b) n Address
n Routing
n Unicast/broadcast/multicast
(a) A switched network
(b) Interconnection of networks

1
 Chapter 1
Cost-Effective Resource Sharing
n FDM: Frequency Division
Multiplexing
n Statistical Multiplexing
n Data is transmitted based
on demand of each flow.
n What is a flow?

n Packets vs. Messages

n FIFO, Round -Robin,

Priorities (Quality
-of-
Service (QoS))
n Congested?

A switch multiplexing packets from multiple sources onto one

LAN, MAN, WAN
shared n
link n SAN (System Area

1
Networks

1
 Chapter 1
Support for Common Services

n Logical Channels
n Application-to-Application communication path or a
pipe

Process communicating over an

abstract channel

1
 Cha

Reliability challenge

n Network software should hide

(inevitable) errors from applications
n Bits are lost n Bit errors (1 to a 0, and vice
versa) n Burst errors – several consecutive
errors
n Packets are dropped (largely
Congestion) n Links and Node failures n
Messages are delayed

1
 Cha

n Messages are delivered out-of-order n

Third parties eavesdrop
Protocols

n Protocol defines the interfaces between the

layers in the same system and with the layers of
peer system
n Building blocks of a network architecture n Each
protocol object has two different interfaces n

1
 service interface: operations on this protocol n peer-to-
peer interface: messages exchanged with
peer
n Term “protocol” is overloaded
n specification of peer-to-peer interface n module that
implements this interface

1
Internet Architecture

Alternative view of the

1
 Cha

Internet architecture. The

Internet Protocol Graph “Network” layer shown here
is sometimes referred to as
the “sub-network” or “link”
layer.
Description of (Lower) Layers

n Physical Layer
n Handles the transmission of raw bits over a communication link
n Data Link Layer
n Collects a stream of bits into a larger aggregate called a frame
n Network adaptor along with device driver in OS implement the

protocol in this layer

1
n Frames are actually delivered to hosts
n Network Layer
n Handles routing among nodes within a packet-switched
network n Unit of data exchanged between nodes in this layer is
called a packet

The lower three layers are implemented on all network nodes

1
 Cha
Description of (Higher) Layers

n Transport Layer
n Implements a process-to-process channel
n Unit of data exchanges in this layer is called a message n
Session Layer
n Provides a name space that is used to tie together the potentially
different transport streams that are part of a single application
n Presentation Layer
n Concerned about the format of data exchanged between peers
n Application Layer n Standardize common type of exchanges

The transport layer and the higher layers typically run only on
endhosts and not on the intermediate switches and routers

2
 Cha
Encapsulation

High-level messages are encapsulated inside of low-level messages

2
Introduction 1-1818

Cha

2
Internet Architecture

n Does not imply strict layering. The application

is free to bypass the defined transport layers
and to directly use IP or
other underlying networks
n An hour-glass shape – wide at the top,
narrow in the middle and wide at the bottom.
IP serves as the focal point for the
architecture
n In order for a new protocol to be officially included in the
architecture, there needs to be both a protocol specification and
at least one (and preferably two) representative implementations
of the specification n IETF Governance
n “rough consensus and running code”
Cha

2
 [email protected]

USE YOU CORNELL EMAIL!

Cha
Benefits of Hourglass architecture
n Internet designed to operate over different underlying
communications technologies, including those yet to be
introduced, and to support multiple and evolving
applications and services.

2
n Does not impede or restrict particular applications (although
users, ISPs may make optimizations)
n Enables developers to write applications without
knowing/adapting to details of underlying networks
n Enables users to adopt applications without involvement/approval
from network operators
n Critical separation between network technology and higherlevel
services through which users actually interact with the
Internet visualized as hourglass

IP as minimal viable agreement/min common

n

denominator maximizes flexibility

2
 Cha
systems, products …

n Tim O’Reilly:
n do as little as possible….the less you include the easier it will be to agree
and you donttie yourself down…because we dontknow what will come
[sic: in this case less is more] n Build a system and let it evolve n Create
architecture for participation—iTUNES, App Store…[sic: it started with
Internet, Includes maps mashups and APIs!!]
n TBL didnthave to ask anyonespermission to put up WWW on the net…
they would have said no…’http is poorly designed protocol..willnever
Why a narrow waist is important metaphor for new

2
 scale’
n Tolerate as much failure and participation as needed to introduce
new systems/innovations rapidly/iteratively and innovate n
Naughton:
n Allow innovation to be tried for free

2
 Cha
End-to-end architecture
n Edge-based innovation derives from early design decision that the Internet
should have an end-to-end architecture:
n The network provides communications fabric connecting the many
computers at its ends
n Network offers very basic level of service, data transport n Beyond
transporting data—locate special features needed to support specific
applications in or close to applications/devices at network edge. n Only put
feature lower down if performance improvement justifies it

n E2E design facilitates n designing for: failure, change, dynamics,

decentralized control, rolling asynchronous adoption, of
components
n scalability and therefore longevity of architecture

3
 Cha
Scalability
n Internet’s design enabled it to support a growing amount of
communications:
n Growth in number of users and attached devices

n n Growth in volume of communications per device and total

n Scale implies…heterogeneity…designing for Heterogeneity is a good

step in future proofing

n QUESTION –Reliable transport of data –Packet level hop by hop,

Packet level end to end (Process), Message/File level end to end
(Application).

3
 Cha

Performance

nLatency = Propagation + transmit + queue n

Propagation = distance/speed of light

n Transmit = size/bandwidth

nOne bit transmission => propagation is important n

Large bytes transmission => bandwidth is important

3
 Cha

Bandwidth

n Width of the frequency band

n Number of bits per second that can be
transmitted over a communication link
n1 Mbps: 1 x 106 bits/second = 1x220
bits/sec
-6
n1 x 10 seconds to transmit each bit or

imagine that a timeline, now each bit occupies 1 micro

second space.

Introduction 33
 Cha

Four sources of packet delay

transmission
A propagation

n On a 2 Mbps link the width is 0.5 micro

second. B
nodal
n Smaller the width more will be transmission
processing queueing
per unit time.
dnodal = dproc + dqueue + dtrans + dprop
(1) dproc: nodal processing § § typically < msec
check bit errors (2) dqueue: queueing delay
§ determine output link

3
§ time waiting at
output link for
transmission
§ depends on
congestion level of
router

Introduction 35
 transmission
Cha
A propagation

(3) dtrans: transmission

B § d: length of physical link
nodal § s: propagation speed in
delay: processing medium (~2x108 m/sec)
queueing
§ L: packet length
prop
(bits) § R: link d
bandwidth (bps) nodal= dproc + dqueue+ dtrans+ dprop
§ dtrans = L/R
(4) dprop: propagation
delay:
d and d § dprop = d/s very different

Introduction 29
 C
Queueing delay -- Congestion
average queueing

delay
n R: link bandwidth (bps)
n L: packet length (bits)
n a: average packet
arrival rate
traffic intensity
= La/R

v La/ ~ 0: avg. queueing delay La/R ~ 0

v R
La/Rsmall
-> 1: avg. queueing delay
v La/R> 1: more
large “work” arriving
than can be serviced, average delay infinite!
 Cha

Packet loss

n queue (aka buffer) preceding link in buffer has finite

capacity
n packet arriving to full queue dropped (aka lost) n lost
packet may be retransmitted by previous node, by
source end system, or not at all

buffer

Introduction 31
 (waiting area
) packet being transmitted
A

B
packet arriving to
full buffer islost
 Chapter 1
Relationship between bandwidth and latency

Note: assuming cross

-country propagation delay
100=ms

1-
3232

A 1-MB file would fill the 1-Mbps link 80 times,

but only fill the 1-Gbps link 1/12 of one time

34
 Cha
Delay X Bandwidth

n We think the channel between a pair of processes as a

hollow pipe
n Latency (delay) length of the pipe and bandwidth the
width of the pipe
n Delay of 50 ms and bandwidth of 45 Mbps Þ 50 x 10-3
seconds x 45 x 106 bits/second Þ 2.25 x 106 bits = 280
KB data.

Network as a pipe

35
 Cha

e.g.,

36
Delay X Bandwidth
n
T

n Relative importance of bandwidth and latency depends on

application
n For large file transfer, bandwidth is critical

n For small messages (HTTP, NFS, etc.), latency is critical n

Variance in latency (jitter) can also affect some applications
( audio/video conferencing) n How many bits the sender
must/could transmit before the first bit arrives at the receiver
akes another one-way latency to receive a response from the receiver
n If the sender does not fill the pipe—send a whole delay × bandwidth
product’s worth of data before it stops to wait for a signal—the sender
will not fully utilize the network
n Control travels over same network as data – latency impairs
feedback which impairs thruput

37
Cha

38