Last Lecture

Nuts-and-bolts description of the Internet

! The topology

! The core
! The edge
! The communication links

SUNY AT BUFFALO; CSE489/589 – MODERN NETWORKING CONCEPTS; Fall 2010; INSTRUCTOR: HUNG Q. NGO 1
 This Lecture

! How to send data from end to end: two switching

methods
! Circuit switching
! Packet switching

! Packet loss and delay in a packet switched network

SUNY AT BUFFALO; CSE489/589 – MODERN NETWORKING CONCEPTS; Fall 2010; INSTRUCTOR: HUNG Q. NGO 2
 How is data transferred through a network?

Two switching methods:

1. Circuit Switching: dedicated physical circuit is established,
maintained, and terminated over a communication session
(e.g. ISDN)

2. Packet Switching: data are transferred in packets (chunks of

data of a fixed size), possibly go through different paths to
reach the destination
(e.g. ATM, X.25, Frame Relay, Internet)

SUNY AT BUFFALO; CSE489/589 – MODERN NETWORKING CONCEPTS; Fall 2010; INSTRUCTOR: HUNG Q. NGO 3
 1. Circuit Switching

Three step process

o Source establishes connection
to destination
" Find path
" Reserve resources
o Data exchanged (no need for
destination address)
o Connection torn down
" Resources released

SUNY AT BUFFALO; CSE489/589 – MODERN NETWORKING CONCEPTS; Fall 2010; INSTRUCTOR: HUNG Q. NGO 4
 Sharing a Link: Multiplexing
! To combine multiple signals (analog or digital) for
transmission over a single line or medium.
! Multiplexing technologies:
! Frequency Division Multiplexing (FDM) : each signal is
assigned a different frequency range (e.g. FM radio).
! Time Division Multiplexing (TDM) : each signal is
assigned a fixed time slot in a “fixed” rotation.
! Statistical Time Division Multiplexing (STDM): time
slots are assigned to signals dynamically to make better
use of bandwidth.
! Wavelength Division Multiplexing (WDM) : each signal
is assigned a particular wavelength; used in optical fiber.

SUNY AT BUFFALO; CSE 489/589 – MODERN NETWORKING CONCEPTS; Fall 2010; INSTRUCTOR: HUNG Q. NGO 5
 Circuit Switching: FDMA and TDMA

Example:
FDMA
4 users

frequency

time
TDMA

frequency

time
SUNY AT BUFFALO; CSE 489/589 – MODERN NETWORKING CONCEPTS; Fall 2010; INSTRUCTOR: HUNG Q. NGO 6
 2. Packet Switching

SUNY AT BUFFALO; CSE 489/589 – MODERN NETWORKING CONCEPTS; Fall 2010; INSTRUCTOR: HUNG Q. NGO 7
 Packet Switching: Statistical Multiplexing

Packets

SUNY AT BUFFALO; CSE 489/589 – MODERN NETWORKING CONCEPTS; Fall 2010; INSTRUCTOR: HUNG Q. NGO 8
 Packet Switching vs. Circuit Switching: In Theory

! Packet Switching
! CS wastes bandwidth when data is sporadic
! PS is statistically more efficient and less costly
! CS takes time to establish the circuit
! PS is simpler to implement
! Side Question: what about packet sizes? Small or Large?
! Circuit Switching
! PS is not suitable for real time application
! A sudden surge of traffic could overflow router’s buffers
! PS could deliver packets in wrong order
! CS is transparent (carrier does not need to know packet
format)
SUNY AT BUFFALO; CSE 489/589 – MODERN NETWORKING CONCEPTS; Fall 2010; INSTRUCTOR: HUNG Q. NGO 12
 Common View of the Telco Network (CS)

brain (smart)

brick (dumb) lock (you can’t get in)

SUNY AT BUFFALO; CSE 489/589 – MODERN NETWORKING CONCEPTS; Fall 2010; INSTRUCTOR: HUNG Q. NGO 13
 Common View of the IP Network (PS)

SUNY AT BUFFALO; CSE 489/589 – MODERN NETWORKING CONCEPTS; Fall 2010; INSTRUCTOR: HUNG Q. NGO 14
 PS vs CS in Practice

Common assumptions about the Internet (That you found in

many textbooks and research papers)
! IP dominates global communications
! Packet switching is more efficient than circuit switching
! Packet switching is robust
! IP (and PS) is simpler
! Quality of Service (QoS) can be realized over IP

SUNY AT BUFFALO; CSE 489/589 – MODERN NETWORKING CONCEPTS; Fall 2010; INSTRUCTOR: HUNG Q. NGO 15
 IP Dominates Global Communications? NO

! [US-census 2002] Revenues: Satellite Telecom (5.7B),

ISPs (18.7B), Radio/TV broadcast (48.5B), Cable Distribution
(77.7B), Cellular & other wireless Telecom (96.5B), Wired
telecom-carriers (237.6B).
! [Nielsen/NetRatings survey 2004 & others]
Percentage of US households having access:
Internet (75%), Cable/Pay TV (78%), TV (98%)
! [RHK Industry Reports 2002] Public Telecom
Infrastructure Expenditures: Core routers (1.7B),
Edge routers (2.4B), SONET/SDH/WDM (28.0B), Telecom
Multi-Service Switches (4.5B)

SUNY AT BUFFALO; CSE 489/589 – MODERN NETWORKING CONCEPTS; Fall 2010; INSTRUCTOR: HUNG Q. NGO 16
 PS is more efficient than CS? Yes, but …
! More efficient means better utilized (both in
transmission lines and switching equipments)
! True for networks with scarce bandwidths
! However, does it really matter today?
! Average utilization levels
! ATT switched voice (33%), Internet backbones (15%)
! Private lines networks (3-5%), LANs (1%)
! Various Reasons
! Internet traffic is asymmetric and bursty, links are symmetric
! Operators tend to over-provision because PS networks behave very
badly once congested (oscillation, routing loops, black holes,
disconnections, etc)
! Over-provision to ensure low delay (satisfy customers), it’s more
economical to add capacity in large increments
SUNY AT BUFFALO; CSE 489/589 – MODERN NETWORKING CONCEPTS; Fall 2010; INSTRUCTOR: HUNG Q. NGO 17
 PS is more robust than CS? Not necessarily …
! Downtime per year:
! Internet: 471min [Labovitz et al. 2000]
! Phone networks: 5min [Kuhn 1997]
! Recover time
! Internet: median 3min, frequently > 15min (due to slow
BGP convergence time)
! SONET/SDH rings: < 50ms (via pre-computed backup
paths)
! Routing in the Internet
! Routing info affected by user traffic, suffering from
congestion (in-band routing)
! Routing computation complex # overload processors
! Probability of mis-configuring a router is high, one
router’s error affect the whole network
SUNY AT BUFFALO; CSE 489/589 – MODERN NETWORKING CONCEPTS; Fall 2010; INSTRUCTOR: HUNG Q. NGO 18
 IP (and PS) is simpler?
! Number of lines of codes in
! Typical Tel. Switches: 3 millions, extremely complex switch: 16M
! Cisco’s IOS: 8 millions [more susceptible to attacks]
! Routers crash frequently, takes long time to reboot
! Hardware
! A line card of a router: OC192 POS has 30M gates + 1 CPU + 300MB
packet buffers + 2MB forwarding table + 10MB other state memory
! Current trend makes routers more complex (multicast, QoS, access
control, security, VPN, etc) – violation of E2E
! A line card of a typical transport switch: ! number of gates, no CPU,
no forwarding table, one on-chip state memory
! Density: highest transport switch capacity = 4 x highest
router capacity, at 1/3 the price
! WDM, DWDM push the difference further
! IP’s “simplicity” does not scale!
SUNY AT BUFFALO; CSE 489/589 – MODERN NETWORKING CONCEPTS; Fall 2010; INSTRUCTOR: HUNG Q. NGO 19
 QoS can be realized over IP?
! Belief: over-provisioning allows low e2e delay #
guaranteeing QoS is possible

! After > 10 years of research, IntServ and DiffServ

are still not good enough.

! Few financial incentive to provide QoS over IP

! Watch out for VoIP, however.
! On the other hand, current phone services are much better
with very low price

SUNY AT BUFFALO; CSE 489/589 – MODERN NETWORKING CONCEPTS; Fall 2010; INSTRUCTOR: HUNG Q. NGO 20
 Other measures
! Scalability
! CS scales more or less linearly
! When data rates increase, routers can’t keep up

! Flexibility
! IP is more flexible
! Lead to high costs of end-systems
! Need more sophisticated users [large organizations need a
room of sys admin, just 1 phone operator]

SUNY AT BUFFALO; CSE 489/589 – MODERN NETWORKING CONCEPTS; Fall 2010; INSTRUCTOR: HUNG Q. NGO 21
 This Lecture

! How to send data from end to end: two switching

methods
! Circuit switching
! Packet switching

! Packet loss and delay in a packet switched network

SUNY AT BUFFALO; CSE489/589 – MODERN NETWORKING CONCEPTS; Fall 2010; INSTRUCTOR: HUNG Q. NGO 22
 How do loss and delay occur?
Packets queued in router buffers
! packet arrival rate to link exceeds output link capacity
! packets queued, wait for turn

packet being transmitted (delay)

B
packets queueing (delay)
free (available) buffers: arriving packets
dropped (loss) if no free buffers

SUNY AT BUFFALO; CSE489/589 – MODERN NETWORKING CONCEPTS; Fall 2010; INSTRUCTOR: HUNG Q. NGO 23
 Four sources of packet delay
! 1. nodal processing:
! check bit errors
determine output link
$ 2. queueing
!

% time waiting at output

link for transmission
% depends on congestion
level of router

transmission
A propagation

B
nodal
processing queueing

SUNY AT BUFFALO; CSE489/589 – MODERN NETWORKING CONCEPTS; Fall 2010; INSTRUCTOR: HUNG Q. NGO 24
 Delay in packet-switched networks
3. Transmission delay:
! R=link data-rate (bps)
4. Propagation delay:
! L=packet length (bits)
! d = length of physical link
! time to send bits into link = L/R
! s = propagation speed in

medium (~2x108 m/sec)

! propagation delay = d/s

Note: s and R are very

different quantities!
transmission
A propagation

B
nodal
processing queueing
SUNY AT BUFFALO; CSE489/589 – MODERN NETWORKING CONCEPTS; Fall 2010; INSTRUCTOR: HUNG Q. NGO 25
 Caravan analogy

100 km 100 km
ten-car toll toll
caravan booth booth
! Time to “push” entire
! cars “propagate” at
caravan through toll booth
100 km/hr
onto highway = 12*10 =
! toll booth takes 12 sec to 120 sec
service car (transmission time)
! Time for last car to
! car~bit; caravan ~ packet propagate from 1st to 2nd
! Q: How long until caravan is toll both: 100km/(100km/
lined up before 2nd toll booth? hr)= 1 hr
! A: 62 minutes
SUNY AT BUFFALO; CSE489/589 – MODERN NETWORKING CONCEPTS; Fall 2010; INSTRUCTOR: HUNG Q. NGO 26
 Caravan analogy (more)

100 km 100 km
ten-car toll toll
caravan booth booth

! Cars now “propagate” at ! Yes! After 7 min, 1st car at

1000 km/hr 2nd booth and 3 cars still at
1st booth.
! Toll booth now takes 1 min to
service a car ! 1st bit of packet can arrive at
2nd router before packet is
! Q: Will cars arrive to 2nd
fully transmitted at 1st
booth before all cars serviced
router!
at 1st booth?
! See Ethernet applet at AWL
Web site

SUNY AT BUFFALO; CSE489/589 – MODERN NETWORKING CONCEPTS; Fall 2010; INSTRUCTOR: HUNG Q. NGO 27
 Nodal delay

! dproc = processing delay

! typically a few microsecs or less
! dqueue = queuing delay
! depends on congestion
! dtrans = transmission delay
! = L/R, significant for low-speed links
! dprop = propagation delay
! a few microsecs to hundreds of msecs
SUNY AT BUFFALO; CSE489/589 – MODERN NETWORKING CONCEPTS; Fall 2010; INSTRUCTOR: HUNG Q. NGO 28
 Queueing delay (revisited)
! R=link bandwidth (bps)
! L=packet length (bits)
! a=average packet arrival rate

traffic intensity = La/R

$ La/R ~ 0: average queueing delay small

$ La/R -> 1: delays become large
$ La/R > 1: more “work” arriving than can be
serviced, average delay infinite!
SUNY AT BUFFALO; CSE489/589 – MODERN NETWORKING CONCEPTS; Fall 2010; INSTRUCTOR: HUNG Q. NGO 29
 “Real” Internet delays and routes
! What do “real” Internet delay & loss look like?
! Traceroute program: provides delay measurement
from source to router along end-end Internet path
towards destination. For all i:
! sends three packets that will reach router i on path
towards destination
! router i will return packets to sender
! sender times interval between transmission and reply.

3 probes 3 probes

3 probes

SUNY AT BUFFALO; CSE489/589 – MODERN NETWORKING CONCEPTS; Fall 2010; INSTRUCTOR: HUNG Q. NGO 30
 Packet loss
! queue (aka buffer) preceding link in buffer has finite
capacity
! packet arriving to full queue dropped (aka lost)
! lost packet may be retransmitted by previous node,
by source end system, or not at all

buffer
(waiting area) packet being transmitted
A

B
packet arriving to
full buffer is lost

SUNY AT BUFFALO; CSE489/589 – MODERN NETWORKING CONCEPTS; Fall 2010; INSTRUCTOR: HUNG Q. NGO 31
 Throughput
! throughput: rate (bits/time unit) at which bits
transferred between sender/receiver
! instantaneous: rate at given point in time
! average: rate over longer period of time

server,
server sendswith link
bits pipe capacity
that can carry link that
pipe capacity
can carry
file of
(fluid) F pipe
into bits Rs bits/sec
fluid at rate Rfluid
c bits/sec
at rate
to send to client Rs bits/sec) Rc bits/sec)

SUNY AT BUFFALO; CSE489/589 – MODERN NETWORKING CONCEPTS; Fall 2010; INSTRUCTOR: HUNG Q. NGO 32
 Throughput (more)
! Rs < Rc What is average end-end throughput?

Rs bits/sec Rc bits/sec

$ Rs > Rc What is average end-end throughput?

Rs bits/sec Rc bits/sec

bottleneck link
link on end-end path that constrains end-end throughput
SUNY AT BUFFALO; CSE489/589 – MODERN NETWORKING CONCEPTS; Fall 2010; INSTRUCTOR: HUNG Q. NGO 33
 Throughput: Internet scenario

! per-connection end-
end throughput is Rs
Rs Rs

Rc Rc
! in practice: Rc or Rs is Rc
often bottleneck

10 connections (fairly) share

backbone bottleneck link R bits/sec
SUNY AT BUFFALO; CSE489/589 – MODERN NETWORKING CONCEPTS; Fall 2010; INSTRUCTOR: HUNG Q. NGO 34