<<

. 42
( 132 .)



>>


• autonomous system
• autonomous system border routers
• areas
• backbone.

Autonomous system
An autonomous system (AS) is a group of routers that exchange routing information within
a single administrative unit. On the Internet, they link to other ASs through autonomous
192 IP APPLICATIONS FOR GPRS/UMTS


system border routers (ASBRs) to allow data to be transferred from one network to
another. (ASs are described in greater detail in Section 5.3.5.1.)

Autonomous system border routers
An autonomous system border router (ASBR) is a router that exchanges routing informa-
tion with routers from other autonomous systems. ASBRs communicate routing informa-
tion with each other through an EGP. ASBRs must be able to translate between the IGP,
e.g. OSPF, and the EGP, e.g. border gateway protocol 4 (BGP4). Figure 5.23 shows how
an ASBR is used to connect to other autonomous systems.

Area
For small or medium-sized networks, distributing data through the internetwork and main-
taining topological databases at each router is not a problem. However, in larger networks
that include hundreds of routers, maintaining the topological database can require several
megabytes of RAM for routing information and heavily utilize the CPU.
For this reason, large networks are often logically divided into smaller networks called
areas. An area usually corresponds to an administrative domain such as a department, a
building or a geographic site. In this way much of the routing information can remain
hidden, thus reducing the burden on the routers. This is shown in Figure 5.24.

Backbone
A backbone is a logical area to which all other areas of the network are connected. This
special area must be directly connected to all other areas of the network (either physically
or virtually).



The Internet


Autonomous System
Border Router



Autonomous System
OSPF




OSPF
OSPF




Figure 5.23 OSPF border router
5.3 IP ROUTING 193



Autonomous System


Area 2
Area 1




Area Border Area Border
OSPF OSPF
Internal
Router Routers
Routers


OSPF OSPF OSPF




Backbone
Area 0



Figure 5.24 OSPF autonomous system connectivity


Routers that attach an area to the backbone are called area border routers (ABRs).
Routers within a particular area receive routing information about the rest of the network
from the ABRs. In this way the internal routers can have their workload reduced. The
ABRs are also responsible for exchanging information about the area with the backbone.
Non-backbone areas can be classi¬ed as either of the following:

• Transit area: an area containing more than one ABR (e.g. area 2 in Figure 5.24).
• Stub area: an area where there is only one ABR (e.g. area 1 in Figure 7.27): all routes
to destinations outside the area must pass through this router.

The backbone routers accept information from the area border routers to compute the
best route from each backbone router to every other router. This information is trans-
mitted back to the area border routers, which advertise it within their areas. Using this
information, a router can select the best route to the backbone for an inter-area packet.
Figure 5.24 shows how the different areas connect together.


5.3.3.2 A comparison of OSPF and RIP
Although it is much more complex, OSPF is considered superior to the RIP routing
protocol for larger networks for the following reasons:
194 IP APPLICATIONS FOR GPRS/UMTS


• RIP can route a packet through no more than 15 routes, since its metric is hop count.
An OSPF metric can be as great as 65 535, thus giving more diversity for routing.
• OSPF networks can detect changes in the internetwork quickly and calculate new routes
faster than RIP, resulting in a faster convergence time. The count-to-in¬nity problem
does not occur in OSPF internetworks.
• OSPF reduces the amount of congestion by generating less traf¬c than RIP. RIP requires
that each router broadcast its entire database every 30 seconds whereas OSPF routers
only broadcast link state information when it changes or every 30 minutes.
• OSPF supports variable-length subnet masks. This allows the network administrator
to assign a different subnet mask for each segment of the network. This increases the
numbers of subnets and hosts that are possible for a single network address. RIP version
2 introduces this feature.

RIP does have advantages over OSPF:

• RIP is a simple protocol and requires little intervention from the administrator; it works
well in small environments.
• Since it is a simpler process, RIP also requires fewer CPU cycles and less memory
than OSPF.



5.3.4 Other routing protocols
There are a number of other routing protocols which are widely used. These are described
brie¬‚y here.


5.3.4.1 Interior gateway routing protocol (IGRP)
IGRP is a distance vector protocol developed by Cisco in the early 1980s. It was devised to
solve the problems associated with using RIP to route datagrams between interior routers.
As such, IGRP converges faster than RIP and does not share RIP™s hop count limitation.
However, like all distance vector protocols, IGRP routers broadcast their complete routing
table periodically, regardless of whether the routing table has changed. IGRP routers do
this by default every 90 seconds. If a router has not received an update for a given path for
180 seconds, it removes the route from its table. This periodic sending of routing table
information wastes bandwidth on the network. IGRP determines the best path through
an internetwork by examining the bandwidth (deduced from the interface type), delay,
reliability and load of the networks between routers.
A more advanced version of IGRP has been developed by Cisco and is known as
enhanced IGRP (EIGRP). It is still a distance vector protocol and uses the same metrics
as IGRP to work out the best route. However, it combines some of the properties of link
state protocols to address the limitations of conventional distance vector routing protocols,
i.e. slow convergence and high bandwidth consumption in a steady-state network. Also,
5.3 IP ROUTING 195


whereas IGRP does not support variable-length subnet masking, EIGRP does support this
function.
Instead of using hop count as a metric, it used various link measurements such as:

• bandwidth on the link;
• delay on the link (not measured but de¬ned as a constant dependent on the technology);
• current load on the link (measured at regular time intervals);
• reliability of the link.

By default generally a Cisco router only uses the ¬rst two of these parameters.
The advantage of using this more complex metric over hop count is that IGRP will work
well if some routes are slower than others, forwarding traf¬c down the faster link. RIP,
for example, is only really well suited to more homogeneous networks where alternate
links have the same bandwidth. IGRP can be used to load balance traf¬c between two
alternative routes to the same destination.
IGRP uses the following techniques to help with convergence:

• split horizon
• poison reverse updates
• triggered updates
• holddowns.

A holddown is where a route is not allowed to be reinstated for a given time after it
has just been removed. This stops the route being reinstated incorrectly by an erroneous
routing table update.


5.3.5 Exterior routing protocols
These have historically been called exterior gateway protocols (EGPs), since gateway is
just another term for router on IP networks such as the Internet. They are used to connect
discrete ASs together. There are two main exterior routing protocols, exterior gateway
protocol (EGP) and border gateway protocol (BGP). EGP has been largely superseded
by BGP, a newer and more versatile protocol. It is rather unfortunate that EGP is the
collective term for these protocols as well as being the term for the speci¬c exterior
gateway protocol. Unlike IGPs, EGPs do not have a single administrative unit and as
such routing decisions are more complicated. Routes may pass through different countries,
each with its own laws on importing and exporting data.


5.3.5.1 Border gateway protocol (BGP)
Routing through an internetwork is relatively straightforward. However, choosing the
best path in a certain set of circumstances can be a dif¬cult problem. EGP was the
196 IP APPLICATIONS FOR GPRS/UMTS


original routing protocol for the Internet and worked satisfactorily in the early days.
However, recently its lack of policy-based route selection has become a big issue. This
is because of the increasingly complicated nature of the Internet, technically, socially,
and politically. BGP was designed to overcome EGP™s problems. Like EGP, BGP is an
inter-AS routing protocol created for use in the Internet™s core routers. As Figure 5.25
shows, BGP enables routing between different ASs which have internally different IGPs,
such as RIP, OSPF or intermediate system“intermediate system (IS“IS). Only the ASBR
needs to understand BGP.
The current EGP on the Internet is BGP4, de¬ned in 1994 and documented in RFC 1771.
BGP requires reliable transportation of updates and as such uses TCP as its transport pro-
tocol, on port 179. When a connection is established, BGP peers exchange complete copies
of their routing tables, which can be quite large. After synchronization only changes are
then exchanged, which makes long-running BGP sessions more ef¬cient than shorter ones.
The primary function of a BGP system is to exchange network reachability information
with other BGP systems (known as BGP speakers). This is achieved through small ˜keep
alive™ packets. If a router does not receive a ˜keep alive™ message from its neighbour
within a certain ˜hold time™ (RFC 1774 suggests these should be sent every 30 seconds)
then it will update its routing table to re¬‚ect the loss of the route. BGP is a path vector
protocol, which has many characteristics that are similar to a distance vector routing
protocol. However, whereas a distance vector routing protocol simply uses the number
of hops to determine the best path, the path vector protocol uses a more sophisticated
policy-based route selection.


Autonomous Autonomous
System System




ASBR ASBR
IGP=IS-IS IGP=OSPF




BGP-4




Autonomous
ASBR
System




IGP=RIP


Figure 5.25 BGP enables connectivity
5.4 TCP AND CONGESTION CONTROL 197


5.4 TCP AND CONGESTION CONTROL

Congestion is caused by network overload. In extreme cases congestion can cause packet
loss due to the over¬‚ow of buffers within the network. This fact is exploited by the TCP
protocol to ensure that congestion is limited within the Internet. Four different algorithms
have been speci¬ed to control congestion for a TCP connection (RFC 2581).


<<

. 42
( 132 .)



>>