Improving the Routing and Addressing of IP

gascitytankΔίκτυα και Επικοινωνίες

28 Οκτ 2013 (πριν από 4 χρόνια και 11 μέρες)

65 εμφανίσεις

1

Improving the Routing and
Addressing of IP

Ford, Rekhter, and Braun

IEEE Network Magazine

May 1993

George Lee

2

Abstract


The current
routing and addressing structure

dose
not adequately scale

to meet the needs generated
by the accelerated growth of the Internet.


This article introduces
hierarchy

into the
interdomain routing system
.


Routing on
IP address prefixes


Summarization


to aggregate multiple routing entries


Renumbering

when site moving


To change host addresses within a site to their new network
service provider


Advantages:


reduction in routing overhead


3

IP Address Structure


32
-
bit long address represented as

dotted quads



Each octet is represented as a decimal number,
separated by dots


Consisting of a
network address

field and a
host
identifier

field


Three

classes

of

unicast


addresses:

4


Subnetting

with a
subnet mask


represented in dotted quad form


1

s: network address


0

s: host identifier


Subnet information is not globally distributed.

5

Routing


IP routing
domain


A collection of hosts and routers under the control of a
single administrative entity using a common routing
system


Intradomain

Routing


Interior Routing Protocols (IGPs)


OSPF (Open Shortest Path First)


IGRP (Interior Gateway Routing Protocol)


RIP (Routing Information Protocol)


IS
-
IS (Intermediate
-
System, Intermediate
-
System)


Interdomain

Routing


Exterior Routing Protocols (EGPs)


EGP (Exterior Routing Protocol)


BGP (Border Gateway Protocol)

6


The relationship between Intradomain and
Interdomain routing can be illustrated by the
NSFNET
:


Hierarchical Architecture


Sites





IGP



EGP




Regional Networks



IGP






EGP





NSFNET Backbone







EGP


Using IGP:


Within Sites


Within Regional Networks


Using EGP:


Between Sites and Regional Networks


Between Sites and NSFNET Backbone


Between Regional Networks and NSFNET Backbone

7


IGP

EGP

8


Hierarchical

structure of NSFNET


Simplifying the routing:


Sites

only need to maintain routing information for their networks
and a
default route

to the regional network


Regional networks

only need to maintain routes to their member
sites and a
default route

to the NSFNET backbone


NSFNET backbone

only need to maintain routes to their
member sites and regional networks


Five top
-
level routing domains:


NSFNET Backbone


Commercial Internet eXchange (CIX)


NASA Science Internet (NSI)


SprintLink


European IP Backbone (EBONE)

9

Current Status of Routing


The current routing of
flat

interdomain routing by
network number
dose not scale

adequately in light
of the Internet

s continued growth.


Memory and computational overhead for routing
information


Bandwidth for routing information distribution


Stability of distributed routing computations


Class B

networks only advertise a single routing
entry (2
14
). But each
Class C

network will require a
separate route entry (2
22
)


resulting in even faster
growth of Internet Interdomain routing system.

10


Hierarchical routing

has desirable
scalable

properties, but requires the use of
hierarchical abstractions

of network
addresses
.


Router software will need to be changed
,
but

these changes can be made
transparent to
host software
.

11

Routing on IP Address Prefixes


Classless Interdomain Rouging (
CIDR
)


IP Address Prefixes (up to 32 bits)


Tuple representation:




<IP network address, bit mask>


bit mask


specifying
contiguous

leading address bits

that are
significant to Internet routing (similar to
subnet mask
)


represented in dotted quad form


Example:


<193.128.0.0, 255.128.0.0>


9
-
bit IP prefixed

(a leading one bit and followed by eight zeros)


100000000

-

11000001.1
0000000.00000000.00000000


Significant network field:
11000001.1

12


Hierarchical Abstraction can be provided by
the
Summarization


A pair of prefixes of length N

can be
summarized to
a single prefix of length N
-
1

if the
prefixes have
the first N
-
1 bits in common
.


Example:

101
0 and
101
1 →
101


Summarization can be
repeatedly applied

to
aggregate

multiple routing entries into a single
entry



Minimizing the number of routes distributed



Scaling better


13


OSPF
,
RIP
-
2
, and
BGP
-
4

can carry 32
-
bit prefix bit
masks.


CIDR was Initially proposed for
Class C

address,
but it can also be used for
Class A and B
.


The current CIDR specifies
three
-
level architecture
,
however, additional levels can be added if needed.


At the
site

level


At the
network service providers


At
continental boundaries


Service providers

may operate on
several regions
,
but they should
obtain addresses

out of the the
continental regions

they serve


to maximize the
summarization capability for each region.

14


Summarization

15

The Best Match is the Longest


The destination address is matched with the
candidate prefixes in the routing table.


The list of prefixes is
sorted by length

and
searched
in the descending order
.


Example: (See the picture on the next page)


Provider A:

<198.1.0.0, 255.255.0.0>

Site S:


<198.1.8.0, 255.255.248.0>



Summarization


A advertises
<198.1.0.0, 255.255.0.0>


Provider B:

<199.3.0.0, 255.255.0.0>

Site T:


<199.3.128.0, 255.255.128.0>



Summarization


B advertises
<199.3.0.0, 255.255.0.0>


Suppose that Site S switches to provider B




B advertises
<199.3.0.0, 255.255.0.0>



16



Summarization




Provider A advertises


<198.1.0.0, 255.255.0.0>




Provider B: advertises


<199.3.0.0, 255.255.0.0>



Routing tables are
sorted by prefix length



When Site S switches to
provider B,


B adds one entry for S



If
B

needs to forward
packet to
Site S
, it will
use
the longest matching
prefix S

instead of prefix
A.

17

Names, Not Addresses, Are Permanent


Renumbering


When a site
switches to a new service provider
,
but
keeps its old address
, additional routing
prefix are advertised.


But, if hosts addresses within that site are
changed to related prefix of the new provider
,
additional routing overhead is reduced.


However, there is not an automatic, easy
-
to
-
use
method for changing host addresses within a
site.

18


1) S

s prefix is part of A

s prefix







2) S switches to provider B

and S

s prefix is till part of A

s
prefix





3) S renumbers into B

s prefix,
Interdomain routing tables shrink

19

Utilization of the IP Address Space


Summarization


A site with
contiguous Class C addresses

can
summarize to
a single routing advertisement


Variable length subnetting


A site can split a Class B network into smaller

area


using prefixes of different length, thus
avoiding the waste on IP addresses

20

Conclusion


To use IP prefixes instead of IP network
addresses


Scale the routing system


Reduce the rate of IP address consumption