Routing is the process of moving packets from one network to
another. The routing decision takes place at the source network
device. That is a router. The decision is made based on metrics used
for a particular routing protocol. Routing protocols may use some or
all of the following metrics in determining the best route to a
- Path length
- Communication cost
Path length is measure in either a cost or a hop count. In
link-state routing protocols, the cost is the sum of the costs
associated with each link in the path. Distance-vector routing
protocols assign a hop count to the path length, which measures the
number of routers a packet traverses between the source and
Reliability is typically the bit-error rate of a link connecting
this router to a source or destination resource. For most of the
routing protocols, the reliability of a link is assigned by the
network engineer. Since it is arbitrary it can be used to influence
and create paths that are favorable over other paths.
The delay metric is an overall measurement of the time it takes
for a packet to move through all the internetworked devices, links
and queues of each router. In addition, network congestion and the
overall distance traveled between the source and destination are
taken into consideration in evaluating the delay metric value.
Because the delay value takes into account many different variables,
it is an influential metric on the optimal path calculation.
Using bandwidth as a metric in optimal path calculations may be
misleading. Though bandwidth of a bandwidth of 1.54 Mbps is greater
than 56 Kbps, it may not be optimal due to the current utilization
of the link or the load on the device on the receiving end of the
The load is a metric that assigns a value to a network resource
based on the resources overall utilization. This value is a
composite of CPU utilization, packets processed per second, and
disassembly/reassembly of packets among other things. The monitoring
of the device resources itself is an intensive process.
In some cases, communication lines are charged based on usage
versus a flat monthly fee for public networks. For example, ISDN
lines are charged based on usage time and potential the amount of
data transmitted during that time. In these instances, communication
cost becomes an important factor in determining the optimal
In designing a routing protocol based network the routing
algorithm should have the following characteristics built into the
Optimality - using some or all of the metrics available for a
routing protocol in order to calculate the optimal route. Different
routing protocols may apply one metric as having a higher weight to
the optimal route calculation than another has. An understanding of
this behavior is important in choosing the routing protocol.
Simplicity - While routing protocols themselves may be
complicated their implementation and operational support must be
simplistic. Router overhead and efficient use of router resources is
important in maintaining a stable and reliable network.
Robustness - Choose a routing algorithm that meets the
requirements of the network design. In some cases, for instance
small networks, a simplistic distance-vector routing protocol is
sufficient. In large networks that require a hierarchical design
requires the ability of the routing protocol to scale to the size of
the network without itself becoming a hindrance on the network.
Rapid Convergence - The convergence time to recalculate and then
use a new optimal path between a source and destination resource is
paramount in meeting availability and service level requirements of
Flexibility - The algorithms employed by the selected routing
protocol must be flexible and adapt to the changing dynamics of
network resources and the network as a whole.
- RIP, RIP2 and IGRP Network Design
RIP, RIP2 and IGRP are distance-vector based routing
protocols. Distance-based vector routing protocols base the
optimal route on the number of hops (i.e., devices) a packet
must pass through to reach a destination. Routing Information
Protocol (RIP) was the first routing protocol algorithm for
distributing, calculating and managing available routes within a
network. Interior Gateway Routing Protocol (IGRP) is a Cisco
proprietary routing protocol algorithm using enhanced optimal
route calculation. IGRP calculates optimal routes based on
bandwidth, delay, reliability and load. RIP2 is the second
generation of RIP. RIP2 supports the Internet Protocol Version 6
specification for 128-bit addressing, variable-length subnet
masks (VLSM) and route summarization.
Distance-vector routing protocols use a flat network
topology as shown in Figure 4.1. Since these protocols are
distance-vector based routing algorithms it is beneficial to
minimize the number of hops between two destinations. This
requires careful planning of the core, distribution and access
topology layers in planning the hierarchical service model.
For most cases, when deploying distance-vector based routing
protocols the service functions of the core, distribution and
access layers typically co-mingle within a single router.
- Addressing and Summarization
In RIP and IGRP networks the IP 16-bit addressing scheme of
IP version 4 is supported. RIP2 supports both the IP version 4
16-bit and IP version 6 128-bit addressing scheme.
Additionally, RIP and IGRP support on fixed subnet masks for a
network. Every subnet address used in the RIP or IGRP network
must use the same subnet masking. RIP2 using VLSM and the
128-bit addressing scheme allows for varied subnet masks of
the router interface. This is because the RIP2 routing packet
includes the subnet mask of the source and destination IP
address. Because RIP2 supports VLSM the routing tables use are
summarized. This reduces the memory requirements on the router
by keeping the routing table to a minimum. RIP and IGRP do not
summarize since every entry represents a unique network or
- Route Selection and Convergence
Both RIP and RIP2 base the optimal route selection on the
number of hops. IGPR enhances this by incorporating bandwidth,
delay, reliability and load. Figure 4.2 illustrates the route
selection difference between RIP, RIP2 and IGRP. RIP and IGRP
use the first route within their routing tables as the optimal
route for a destination network or subnet. RIP does not load
balance so multiple entries within the table for a destination
network only become available if the optimal route is
recalculated as less favorable. IGRP will load balance packets
over equal-cost paths to s destination network or subnet. This
load balancing occurs in a round-robin fashion. Both RIP and
IGRP build their tables and then transmit the entire routing
table to adjacent routers. Each router in turn recalculates
its table based on the information received from the sending
router. Once this is completed the router forwards its new
table to adjacent routers. Both RIP and IGRP periodically send
their routing tables to adjacent routers. RIP defaults to a 30
second interval for sending the routing table to adjacent
routers. IGRP defaults to a 90 seconds interval for sending
the routing table to adjacent routers. Both RIP and IGRP will
recalculate routing entries once recognizing a link outage or
timeout to an adjacent router. However, the recalculated
routing table is not forwarded to adjacent routers until the
update interval has been reached. The periodic updating of
neighbor routers for topology changes causes excessive
convergence time for the network to learn new optimal
RIP2 however, addresses the periodic update problem by
sending only the updated route entry at the time of the
recalculation. While this sounds much like a link-state
protocol update RIP2 still sends the entire table on a
periodic basis. The ability of RIP2 to send an update at the
time it is recalculated reduces the convergence time. RIP2
sends the entire routing table on a periodic basis just as RIP
and IGRP. However, the table is smaller due to the use of VLSM
and route summarization. RIP2 will load balance packets to a
destination network or subnet over equal-cost paths.
- Network Scalability
The time for convergence of RIP, IGRP and RIP2 networks is
the single inhibitor to scaling these protocols to large
networks. Convergence is not just a time factor but also a CPU
and memory issue on each router. These protocols recalculate the
entire table during convergence versus just the affected route.
Therefore, convergence becomes a CPU intensive process thereby
reducing the ability of a router to provide service levels
during convergence. Since these protocols send the entire table
in a periodic timeframe they consume bandwidth causing bandwidth
constraints in an ongoing basis.
- EIGRP Network Design
Enhanced Interior Gateway Protocol (EIGRP) is a proprietary
routing protocol of Cisco Systems. EIGRP merges the best of
distance-vector protocol characteristic with advantages of
link-state protocol characteristics. In addition, EIGRP uses
Diffusing Update Algorithm (DUAL) for fast convergence and
further reduction of possible routing loops with in the network.
An advantage to using EIGRP over other routing protocols is its
ability to support not only IP but also Novell NetWare IPX, and
AppleTalk, thus simplifying network design and
EIGRP uses a non-hierarchical flat networking topology.
EIGRP automatically summarizes subnet router for networks
directly connected to the router using the network number as
the boundary. It has been found that the automatic
summarization is sufficient for most IP networks.
- Addressing and Summarization
EIGRP supports variable-length subnet masking (VLSM).
Defining an address space for use by an EIGRP is a primary
step in developing the routing architecture. EIGRP support for
VLSM is made possible by including the subnet mask assigned to
the router interface in the EIGRP routing messages. VLSM is
essentially the subnetting of a subnet (sub-subnet). Using an
appropriate addressing scheme, the size of the routing tables
and convergence time can drastically be reduced through route
summarization. EIGRP automatically summarizes the routes at
network number boundaries. Figure 4.3 diagrams the use of
route summarization. However, the network engineer can
configure route summarization at the interface level using any
bit-boundary of the address to further summarize the routing
entries. The metric used in route summarization is the best
route found for the routes used to determine the summarized
- Route Selection
EIGRP uses the same metrics as IGRP. These values are
bandwidth, delay, reliability and load. The metric placed on a
route using EIGRP defaults to the using the minimum bandwidth
of each hop plus a media-specific delay for each hop. The
value for the metrics used in EIGRP are determined s
Bandwidth - EIGRP uses the default value for each interface
to the value specified by the bandwidth interface command.
Delay - The inherent delay associated with an interface.
The delay metric can also be defined on an interface using the
delay interface command.
Reliability - A dynamically computed value averaged over
five seconds. The reliability metric changes with each new
Load - A dynamically computed weighted average over five
seconds. The load metric changes with each new weighted
EIGRP employs Diffusing Update Algorithm (DUAL) for
calculating route computations. DUAL uses distance vector
algorithms to determine loop-free efficient paths selecting
the best path for insertion into the routing table. DUAL
however, also determines the second best optimal route for
each entry termed a feasible successor. The feasible successor
entry is used when the primary route becomes unavailable.
Figure 4.4 illustrates the use of the feasible successor.
Using this methodology of successor routes avoids a
recalculation and therefore minimizes convergence time. Along
with primary routes, EIGRP distributes the feasible successor
entries to the neighboring routers.
Scalability is a function of memory, CPU and bandwidth
efficiencies. EIGRP is architected in optimizing these
resources. Through route summarization, the routes advertised
by neighbors are stored with minimal memory required. This
enables an EIGRP network to expand without routing issues.
Since EIGRP uses DUAL only routes that are affected by a
change are recomputed and since EIGRP is based on using the
same metrics as IGRP the computation CPU requirements are
minimal. Because EIGPR only sends updates due to topology
changes bandwidth is preserved. Steady-state bandwidth
utilization of EIGRP is minimal due to the use of EIGRP's
HELLO protocol for maintaining adjacencies between
Since EIGRP is a Cisco IOS proprietary routing protocol it is
available only on Cisco routers. Additionally, route filters and
authentication can be specified to further limit accidental or
malicious routing disruptions from unknown routers connecting to
- OSPF Network Design
Open Shortest Path First (OSPF) is a standards based
link-state routing protocol defined by the Internet Engineering
Task Force (IETF) OSPF workgroup and published in Request for
Comment (RFC) 1247. OSPF is based on autonomous system (AS).
OSPF defines an AS as a group of routers exchanging routing
information using link-state protocol. OSPF is based on using a
hierarchical networking topology. Defining the hierarchy
requires planning to define boundaries that denote an OSPF area
and address assignment.
OSPF defines its hierarchy based on areas. Figure 4.5
illustrates the OSPF hierarchy and various areas used to build
and connect the OSPF network. An area is a common grouping of
routers and their interfaces. OSPF has one single common area
through which all other areas communicate. Due to the use of
the OSPF algorithm and its demand on router resources it is
necessary to keep the number of routers at 50 or below per
OSPF area. Areas with unreliable links will therefore require
many recalculations and are best suited to operate within
The OSPF algorithm using a flooding technique for notifying
neighbors of topology changes. The greater number of neighbors
the more CPU intensive the topology change since the new route
must be recalculated and forwarded to all attached neighbors.
Cisco studies have resulted in a recommendation of no more
than 60 neighbors per OSPF router.
The OSPF link-state algorithm calculates a change for each
specified area defined on the router. Area routers are usually
also area border routers (ABR). That is they maintain and
support OSPF routing tables for two OSPF areas. In general,
there is a minimum of two areas for an ABR: The backbone area
and one non-backbone area. The recommendation for OSPF is to
limit the number of supported areas in a router to three. This
will minimizes resources utilization for the calculation and
distribution of link-state updates.
OSPF uses a designated router as the keeper of all the OSPF
routes within a local-area network. This reduces routing
updates over a LAN thereby preserving LAN media bandwidth.
OSPF routers attached to the same LAN as the designated router
request a route only if their own table does not have an entry
for the destination resource. A backup designated router is
also used for availability and redundancy. The recommendation
is to have a designated and backup designated router
supporting only one LAN. In addition, the designated and
backup designated router should be the least CPU intensive
router on the LAN.
The OSPF backbone must be designed for stability and
redundancy. A link failure that partitions the backbone will
result in application outages, which leads to poor
availability. The size of the backbone should follow that
recommended areas to be no more than 50 routers.
Routers within the OSPF backbone must be contiguous. This
follows the concept of the hierarchy and maintains the traffic
for backbone updates within the backbone area routers.
However, OSPF offers the use of a virtual link for connecting
two non-contiguous routers through a non-native area router.
Using a virtual link, a partitioned backbone can be
circumvented until the link failure causing the outage is
corrected. Finally, reserve the media used for the OSPF
backbone for routers to avoid instability and unrelated
routing protocol traffic.
As with backbone areas each OSPF area must be contiguous.
Not only contiguous in design but also contiguous in the
network address space. Using a contiguous address space makes
route summarization possible. The routers of an area
connecting the area to the OSPF backbone area are termed area
border routers (ABR). For availability, it is deemed
appropriate to have more than one ABR connecting the area to
the backbone area.
Designing large-scale OSPF networks requires a review of
the physical connectivity map between routers and the density
of resources. Designing the network into geographic areas may
be beneficial for simplifying implementation and operations
but may not be beneficial for availability or performance. In
general, smaller OSPF areas generate better performance and
higher levels of availability than large OSPF areas.
- Addressing and Summarization
Maximizing the address space in OSPF networks assists in reducing
resource utilization and maximizes route summarization. A
hierarchical addressing scheme is the most effective means of
designing an OSPF network. OSPF supports VLSM that lends itself to a
hierarchical network address space specification. Using VLSM, route
summarization is maximized at the backbone and ABR routers.
Guidelines in defining an OSPF network for optimized route
- Define the network address scheme in subnet ranges for use in
each contiguous area.
- Use VLSM addressing to maximize address space.
- Define the network address space for future growth to allow
the splitting of an area.
- Design the network with the intention of adding new OSPF
routers in the future.
Route summarization increases the stability of an OSPF network.
Using route summarization keeps route changes within an area. Route
summarization must be explicitly specified when working with OSPF
networks on Cisco routers. The specification of router summarization
requires the following information:
- Determine route information needed by the backbone about each
- Determine route information needed by an area for the backbone
and other areas
OSPF route summarization occurs in area border routers. Using
VLSM, bit-boundary summarization is possible on network or subnet
addresses within the area. Since, OSPF route summarization is
explicit the network design must incorporate summarization
definitions for each OSPF area border router.
OSPF areas offer four types of routing information. These
Default - A default route of all packets for which the
destination IP network or subnet is not explicitly found in the
Intra-area routes - These are routes for network or subnets
within a given area.
Interarea routes - This information provides areas with explicit
network or subnet routers for networks or subnets within the OSPF
autonomous system but not within the area.
External routes - These are routes learned from the exchange of
routing information between autonomous systems. This results in
routes that are external to the OSPF autonomous system.
OSPF route information provides information on three types of
OSPF areas. These are non-stub areas, stub areas and stub areas
without summaries. Stub areas are OSPF areas that connect only to
one other area and therefore are considered a stub off the
hierarchy. A non-stub area is an OSPF area that provides
connectivity to more than one OSPF area.
Non-stub area characteristics are:
- Store default routes, static routes, intra-area routes
interarea routes and external routes.
- OSPF interarea connectivity.
- Uses autonomous system border routers.
- Virtual links require non-stub areas.
- Most resource-intensive type of area.
Stub area characteristics are:
- Build default, intra-area, and interarea routes.
- Most useful in areas containing one ABR
- May contain multiple area border routers to same area
- Virtual links cannot connect through stub areas
- Cannot use autonomous system border routers.
Stub areas without summaries contain:
- Default and intra-area routers
- Recommended for single router connections to the
Table 4.x lists the OSPF area types against the routing
Routing Information type
Stub without summaries
- Route Selection
OSPF defaults route selection to the bandwidth metric.
Under OSPF the bandwidth metric is determined by the type of
media being used. The bandwidth metric for a link is the
inverse of the bandwidth supported by the media used for the
link. The bandwidth metric has been calibrated based on a
metric of 1 for FDDI media. Figure 4.6 depicts an OSPF network
and the applied bandwidth metric. The total metric for a given
route is the sum of all the bandwidth metric values of all the
links used for the route. Media that supports bandwidth
greater than FDDI 100 Mbps default to the FDDI metric value of
1. In a configuration where media types connecting the router
are faster than FDDI a manual cost greater than 1 must be
applied to the FDDI link in order to favor the higher speed
OSPF route summarization uses the metric of the best route
found within the summarized routes as a metric value for the
OSPF external routes are defined as being either a type 1
or type 2 route. The metric for a type 1 external route is the
sum of the internal OSPF metric and the external route metric.
Type 2 external routes use only the metric of the external
route. Type 1 external route metrics are more favorable in
providing a truer metric for connecting to the external
For single ABR OSPF areas, all traffic leaving the area
flows through the single ABR. This is done by having the ABR
exchange a default route with the other routers of the area.
In multiple ABR OSPF areas, the traffic can leave either
through the ABR closest to the source of the traffic or the
ABR nearer to the destination of the traffic. In this case,
the ABRs exchange summarized routes with the other routers of
High availability network design requires redundant paths
and routers. Redundancy is useful when employing equal-cost
paths to take advantage of load balancing. Cisco routers will
load-balance over a maximum of four equal-cost paths between a
source and destination using either per-destination or
per-packet load balancing when using OSPF. The default of
per-destination is based on connectivity bandwidth at 56 Kbps
Since OSPF is a link-state based routing protocol, it
adapts quickly to network topology changes. OSPF detects
topology changes based on interface status or the failure to
receive a response to an OSPF HELLO packet of an attached
neighbor within a given amount of time. OSPF has a default
timer of 40 seconds in broadcast networks (i.e., LANs) and two
minutes in non-broadcast networks (i.e., WANs).
The routes are recalculated by the router recognizing the
failed link and sends a link-state packet to all the routers
within the area. Each router then recalculates all the routes
within its routing table.
The addressing scheme, number of areas and number of links
within the OSPF network all affect the scalability of an OSPF
network. Routers use memory for storing all the link states
for each area a router belongs. The more areas attached to a
router the larger the table. Scaling OSPF therefore depends on
the effective use of route summarization and stub areas to
reduce memory requirements. The larger the link-state database
the more CPU cycles required during recalculation of the
shortest path first algorithm. Minimizing the size of a OSPF
area and the number of links within the area along with route
summarization enables OSPF to scale to large networks. OSPF
only sends small HELLO packets and link-state updates when a
topology change occurs or at start-up. This is a great benefit
for preserving bandwidth utilization as compared to
distance-vector routing protocols such as RIP or IGRP.
OPSF can use an authentication field to verify that a router
connecting as a neighbor is indeed a router that belongs within the
network. OSPF routers by their very nature do not allow the
filtering of routes since all OSPF routers must have the same
routing information within an area. Using authentication, an OSPF
router can verify that it should exchange topology information with
a new router that has joined the network. In this way, not only does
OSPF provide some protection from unwanted access, it assists in
keeping a stable network.