Transcript Conclusions and project presentation - comp

1
A COURSE-END
CONCLUSIONS
Dr. Rocky K. C. Chang
The internetworking problem
2

The internetworking problem
 Different
data-link protocols
 Different MAC address spaces
 Different MTUs

An hour-glass model (end-to-end argument)
 IP
as the glue
 IP addresses
 IP fragmentation and reassembly

IP over anything and everything
All boil down to the design goals
3

Best effort IP service
 Reliability
requirement: trusting states to the end hosts
 Requirement for supporting a variety of TOS
 Requirement for accommodating a variety of networks

End-to-end argument
 Keep
the network simple: IP packets go in and IP
packets come out.
 Implement applications at the network edge.
The forwarding problem
4

Main issues of concerns
Who knows what
 Speed (forwarding table size and lookup)
 Not responsible for the correctness of the routes


Hop-by-hop forwarding as a result of the best-effort
approach.
Source routing and tunnels
 Virtual circuit switching
 IP switching


From classful to classless routing
The routing problem
5



THE intelligence of the IP layer
Use a hop-by-hop protocol to deliver packets end-toend.
Main issues of concerns
Speed of convergence
 Prone to routing loops
 Efficiency


Two main approaches (in midst of many other
differences and variations)

Distance vector and link state
A tale of two routing problems
6

All routing protocols concern delivering packets
from one point to another.
 An
intradomain routing additionally concerns optimizing
certain costs of a route.
 An interdomain routing additionally concerns satisfying
certain policies of an AS.

Current Internet characteristics
 Asymmetric
routes
 Packet reordering
 Packet losses
 Nonfriendly intermediaries
The end-to-end problems
7

TCP adds the following services to IP:
 Multiplexing
(through the port number)
 Inordering (through the TCP SN)
 At-most-one-copy (through the TCP SN)
 Arbitrarily large application messages (through the
wraparound TCP SN space)
 Flow control (through advertised window)
 End-to-end reliability (through the sliding window
protocol and retransmission)
 Congestion control (through ACK clocking, congestion
window, slow start, etc)
The congestion control problem
8


Congestion control and/or resource allocation hold one of the
keys to the Internet stability.
A TCP sender interprets packet losses (without receiving ACKs)
as a sign of congestion.



Slow starting to trigger packet losses (reaching the network capacity)
Next time, perform congestion avoidance when approaching to the
congestion point.
Other approaches do not induce packet losses.

TCP/Vegas, Explicit Congestion Notification
Two Internet applications
9

DNS provides
a distributed database for domain names and
 protocols to obtain their resource records.


Web provides
A global naming system to identify resources
 A text-based language to facilitate a navigation across
various related resources, and
 A protocol for requesting and responding



Interaction between TCP and HTTP
Web proxies: not longer end-to-end
Coverage in terms of protocols
10
Application
Transport
Network
BGP
FTP
Telnet
RIP-I/II
SMTP
DNS
TLS
RTP
TCP
UDP
IPSec
IGMP
PIM
OSPF
DVMRP
ICMPv4
IPv4
IPv6
ICMPv6
ARP
HTTP1.0
/1.1
IKE
DHCP
RSVP
Mobile
IPv4
Mobile
IPv6
RTSP
SIP