Vous êtes sur la page 1sur 31

Computer Networks

ENGG1015
1st Semester, 2010
Hayden Kwok-Hay So
Where are we in the semester?
High Applications •  Image & Video Processing
Level
•  Computer & Embedded Systems
Systems •  Computer Network
•  Mobile Network

Digital •  Combinational Logic


Logic •  Boolean Algebra

Circuits •  Basic Circuit Theory

Low Electrical •  Voltage, Current


Level Signals •  Power & Energy

1st semester, 2010 ENGG1015 - Network - Dr. H. So 2


Introduction
  Computer Network:
•  A collection of communicating computers

1st semester, 2010 ENGG1015 - Network - Dr. H. So 3


Categorizing Computer Networks
  LAN
•  Network that covers a small geometrical area.
•  A company, a university, home, etc…
  WAN
•  Network that covers a relatively larger geometrical
area.
•  Usually beyond a building, a city, or even a
country
•  Global network?
•  Beyond the Earth?
  Many other ways to categorize networks

1st semester, 2010 ENGG1015 - Network - Dr. H. So 4


Direct Connections
Computer C Number of links proportional to Computer D
number of computers

Direct (Physical)
Connection
Computer A Computer B

1st semester, 2010 ENGG1015 - Network - Dr. H. So 5


Switching Network
Computer C Computer D

switch

Must share the physical


Computer A connection to A to support data Computer B
connection to/from C, D

1st semester, 2010 ENGG1015 - Network - Dr. H. So 6


How to share resources (switch)?
  Twoimportant ways of network
communication:
  Circuit switching
•  dedicated circuit per connection
•  e.g. public switch telephone network
(PSTN)
  Packet switching
•  Data sent through the network in discrete
packets
•  e.g. Internet

1st semester, 2010 ENGG1015 - Network - Dr. H. So 7


Circuit Switching
End-to-end resources reserved for exclusive use by
each connection
  Network resources (e.g.,
bandwidth) divided into “pieces”
  Pieces allocated to each
connection
  Guaranteed performance
  Resource piece idle if not used
by owning call (no sharing)
  How to divide resources?
•  frequency division
•  time division

1st semester, 2010 ENGG1015 - Network - Dr. H. So 8


Packet Switching
10 Mbs
A Ethernet statistical multiplexing C

1.5 Mbs
B
queue of packets
waiting for output
link
D E
  Each end-end data stream divided into packets
  Different users share the same resource as needed
  Excessive queuing might cause packet drop

1st semester, 2010 ENGG1015 - Network - Dr. H. So 9


Packet switch vs circuit switch
  1 Mbit link
  each user: 1 Mbps link
•  100 kbps when active N users
•  active 10% of time

  circuit-switching:
•  10 users
  packet switching:
•  with 35 users, probability that > 10
users active is less than .0004

Packet switching allows more users to use network!

1st semester, 2010 ENGG1015 - Network - Dr. H. So 10


Packet Implementation
Header Payload
  Two main parts of a packet:
•  Header
•  Payload
  Header:
•  Meta information about the content of the packet
•  e.g. Source and destination address, type of packet, length
etc
  Payload:
•  The actual content of the packet
•  Length of payload specified in header, or it might be a fixed
length
•  In case of fixed length payload: padding is needed

1st semester, 2010 ENGG1015 - Network - Dr. H. So 11


Packet Routing
  The header of a packet states should
contains enough information for the
delivery of a packet.
  The processing of determining and
delivering packets is called routing.
  Two common ways to route:
•  Source routing
•  Multi-hop routing

1st semester, 2010 ENGG1015 - Network - Dr. H. So 12


Source Routing
  Before a packet is sent, the source has
already figured out the complete route the
destination
  Storethe calculated route in the header of
a packet
  Similar to a direction to a location:
•  Turn left at the 1st traffic light, then
•  Go straight, then
•  Turn right at the postoffice …
  No address is needed.

1st semester, 2010 ENGG1015 - Network - Dr. H. So 13


Multi-hop Routing
  Onlyneed destination address in the
packet
  Machines along the way determine how
to deliver the packet to the destination
  Each computer-computer transfer is
referred as a hop.
  Usually designated machines are used
only for routing purposes, we call them
routers.

1st semester, 2010 ENGG1015 - Network - Dr. H. So 14


Packet Forwarding Techniques
A B

  Packets must travel from point A to


point B via many intermediate points
  How should each intermediate point
handle the packet?
  Two common switching methods:
•  Store-and-forward
•  Cut-through

1st semester, 2010 ENGG1015 - Network - Dr. H. So 15


Store-and-Forward
A B

Packets are forwarded out only after the


entire packet has been received

  Easy to implement   Need big buffers


  Allow error checking   Transfer latency
at each intermediate proportional to the
steps number of hops

1st semester, 2010 ENGG1015 - Network - Dr. H. So 16


Store-and-Forward Performance
L
R R R

Takes L/R seconds to Example:


transmit (push out)
packet of L bits on to   L = 7.5 Mbits
link of R bps   R = 1.5 Mbps
  Entire packet must
arrive at router before it   delay = 15 sec
can be transmitted on
next link
  delay = 3*L/R

1st semester, 2010 ENGG1015 - Network - Dr. H. So 17


Packet Switching: Message Segmenting
Now break up the message into
5000 packets

  Each packet 1,500 bits


  1 msec to transmit
packet on one link
  pipelining: each link
works in parallel
  Delay reduced from 15
sec to 5.002 sec

1st semester, 2010 ENGG1015 - Network - Dr. H. So 18


Cut-Through Forwarding
  Packets are forwarded as soon as its
destination address is known
•  Destination address preferred to be at the front of
a packet
  Cannot perform error checking
•  Error can only be detected once the entire packet
has arrived at the router.
  No need for large packet buffers
  Can cause queue-blocking or even deadlock
in some networks
  Lower latency than store-and-forward, but
throughput the same.

1st semester, 2010 ENGG1015 - Network - Dr. H. So 19


Is packet switching always better?
  Great for bursty data
•  resource sharing
•  simpler, no call setup
  Excessive congestion: packet delay and loss
•  protocols needed for reliable data transfer,
congestion control
  Q: How to provide circuit-like behavior?
•  bandwidth guarantees needed for audio/
video apps
•  still an unsolved problem

1st semester, 2010 ENGG1015 - Network - Dr. H. So 20


Announcements
  Homework 3 is online now
•  Due in 2 weeks: 11/24
  18/11:
•  Lecture venue will be moved to CYC-A
  1/12:
Final project presentation +
competition
  20/12: Final exam
•  Reference textbooks at
http://www.eee.hku.hk/~engg1015/fa10/
links.html

1st semester, 2010 ENGG1015 - Network - Dr. H. So 21


1st semester, 2010 ENGG1015 - Network - Dr. H. So 22
Measuring Network Performance

Latency Throughput

  Measures the time   Measures the


to get the first byte aggregated amount
of data across a of data transported
network over a period of time
  Unit: seconds (s)   Unit: byte per
second (Bps) or bit
per second (bps)

1st semester, 2010 ENGG1015 - Network - Dr. H. So 23


Example: Mailing DVDs
  Mailing a DVD takes 1 day
•  = 24 hours = 86400s
  Each DVD is 4GB
   Latency is 86,400s
  Throughput is an average value.
  Ifwe mail 1 DVD everyday, then the
throughput is:
•  4GB / 86400s ≈ 46.2 kBps ≈ 370 kbps

1st semester, 2010 ENGG1015 - Network - Dr. H. So 24


Example: scaling up
  What if we mail 100 DVDs at the same time?
  Latency remains the same:
•  86400s
  Bandwidth increases by 100 times
•  ≈ 37 Mbps
  In other words, mailing 100 DVDs at the same
has higher throughput than your average
“broadband” internet access at home.

1st semester, 2010 ENGG1015 - Network - Dr. H. So 25


Physical Mail vs Network
  When the data transfer is large, traditional postal mail has
much higher bandwidth than any Internet connection
  E.g. Amazon AWS Import/Export allows company to send
data to the Internet “cloud” by physically sending hard
drives.
  When is physical mail faster?
Available Internet Theoretical Min. When to Consider
Connection Number of Days to AWS Import/Export?
Transfer 1TB at 80%
Network Utilization
T1 (1.544Mbps) 82 days 100GB or more
10 Mbps 13 days 600GB or more
T3 (44.736Mbps) 3 days 2TB or more
100 Mbps 1-2 days 5TB or more
1000 Mbps < 1day 60TB or more

1st semester, 2010 ENGG1015 - Network - Dr. H. So 26


Which one to use?
  Both are important
  Latency is more important when small amount
of data is needed to be sent back-and-forth
very quickly
•  Interactive applications such as:
•  Web surfing, online game, remote log in, etc
•  Microprocessor network
  Throughput is more important when streaming
data are considered
•  Video streaming
•  Music streaming
•  Large file transfer

1st semester, 2010 ENGG1015 - Network - Dr. H. So 27


Which network is better?
time

time

time

  Network A has low latency and low throughput


  Network B has high latency, high bandwidth
  Network C has moderate latency, moderate
throughput, and unpredictable performance

1st semester, 2010 ENGG1015 - Network - Dr. H. So 28


Which network is better? (A)
time

time

time

  Network A is the best for short, bursty


traffic, such as web browsing.
•  Think of the first yellow packet as all you need
for 1 web page
  Email, IM (MSN messenger, gtalk,etc) are
all examples of short bursty traffic
1st semester, 2010 ENGG1015 - Network - Dr. H. So 29
Which network is better? (B)
time

time

time

  Network B has the longest latency (longest wait


time)
  But once the data arrive, it has the highest
bandwidth
  Network B is best for streaming data
  E.g. youtube, and any other streaming data

1st semester, 2010 ENGG1015 - Network - Dr. H. So 30


Which network is better? (C)
time

time

time

  Network C is difficult to classified


  Network C is the most realistic model
of today’s ISP’s capability

1st semester, 2010 ENGG1015 - Network - Dr. H. So 31

Vous aimerez peut-être aussi