This chapter discusses the following advanced IP addressing topics:
IP Address Planning
Hierarchical Addressing Using Variable-Length Subnet Masks
Classless Interdomain Routing
Network Address Translation
Understanding IP Version 6
Scalable, well-behaved networks are not accidental; they are the result of good network design and effective implementation planning. A key element for effective scalable network implementation is a well-conceived and scalable IP addressing plan. The purpose of a scalable IP addressing plan is to maximize the amount of IP address space available in deployed networks (this address space is shrinking) and to minimize the size of routing tables.
As a network grows, the number of subnets and the volume of network addresses required increase proportionally. Without advanced IP addressing techniques such as summarization and classless interdomain routing (CIDR), the size of the routing tables increases, which causes a variety of problems. For example, networks require more CPU resources to respond to each topology change in the larger routing tables. In addition, larger routing tables can cause delays while the CPU sorts and searches for a match to a destination address. Both of these problems are solved by summarization and CIDR.
To effectively use summarization and CIDR to control the size of routing tables, network administrators employ other advanced IP addressing techniques such as Network Address Translation (NAT) and variable-length subnet masking (VLSM).
NAT allows the use of a private addressing space within an organization while using globally unique addresses for routing across the Internet and between independent divisions of the organization. Different address pools may be used to track groups of users, which makes it easier to manage interconnectivity.
VLSM allows the network administrator to subnet a previously subnetted address to make the best use of the available address space.
Another long-standing problem that network administrators must overcome is the exhaustion of available IP addresses caused by the increase in Internet use. Although the current solution is to use NAT, the long-term solution is to migrate from the IP version 4 (IPv4) 32-bit address space to the IP version 6 (IPv6) 128-bit address space. Gaining insight into IPv6 functionality and deployment will prove valuable for network administrators in the not-too-distant future.
After completing this chapter, you will be able to describe the concepts of network design and explain the benefits and characteristics of an effective scalable IP addressing plan. You will also be able to describe the role of VLSM addressing in a scalable network and calculate VLSM addresses for a network. You will be able to demonstrate the principles of route summarization and CIDR by summarizing a given range of network addresses into larger IP address blocks. You will also be able to configure NAT for multiple address pools using access lists and route maps. Finally, you will be able to describe the features and benefits of using IPv6.
IP Address Planning
A well-designed large-scale internetwork with an effective IP addressing plan has many benefits. It is scalable, flexible, predictable, and can reduce the routing table size through summarization.
Scalable Network Design
An understanding of scalable network design concepts is imperative for understanding proper IP address planning.
Corporate organizational structure should affect network design. The structure of a scalable network design reflects a corporation's information flow and is called a hierarchical network design.
There are two types of hierarchical network design: functional and geographic.
The design concepts discussed in this section are only a very small part of good network design from the perspective of the IP addressing plan. For a full discussion of internetwork design, refer to CCDA Self-Study: Designing for Cisco Internetwork Solutions (DESGN) (Cisco Press, 2003).
Functional Structured Design
Some corporations have independent divisions that are responsible for their own operations, including networking. These divisions interact with one another and share resources; however, each division has an independent chain of command.
This type of corporate structure is reflected in a functional network design, as illustrated in Figure 1-1. In this example, the different divisions of the corporation have their own networks and are connected according to their functional purpose within the corporate structure. The network architecture can follow the corporate organizational chart.
Figure 1-1 In a Functional Design, Networks Are Connected According to Their Functional Purpose
Geographic Structured Design
Many retail corporations are organized by the geographic location of their stores. Within the corporate structure, each local retail store reports to a district consolidation point. These district consolidation points report to regional consolidation points; the regional consolidation points then report to corporate headquarters. Networks are organized along geographic boundaries, such as countries, states, or provinces.
This type of corporate structure is reflected in a geographic network design, as illustrated in Figure 1-2. In this example, the divisions of the corporation have their own networks and are connected according to their location.
Figure 1-2 In a Geographic Design, Networks Are Connected According to Their Location
From a networking point of view, a geographic network structure is cost-effective because fewer network links require long-haul carriers, often a considerable added expense.
Within the functional or geographic networks, the following three primary layer elements are involved in a hierarchical scalable network design:
Access layerProvides local and remote workgroup, end-user, and customer access to the network. Virtual LANs (VLANs), firewalls, and access lists maintain security for this layer.
Distribution layerProvides policy-based connectivity and is the consolidation point for access layer devices and corporate services. Host services required by multiple access layer devices are assigned to this layer.
Core (or backbone) layerProvides high-speed transport to satisfy the connectivity and transport needs of the distribution layer devices. The circuits with the fastest bandwidth are in the core layer of the network. Redundancy occurs more frequently at this layer than at the other layers.
There are many different ways of designing these hierarchical layers. Some of the considerations are identified in this section.
Fully Meshed Core Layer
The core layer is designed to provide quick and efficient access to headquarters and other divisions within a company. Because the core is usually critical to the network, redundancy is often found in this layer. In a fully meshed core layer design, shown in Figure 1-3, each division has redundant routers at the core layer. The core sites are fully meshed, meaning that all routers have direct connections to all other routers. This connectivity allows the network to react quickly when it must route data flow from a downed link to another path.
Figure 1-3 In a Fully Meshed Core, All Routers Are Connected to All Other Routers
For a small core with a limited number of divisions, this core layer design provides robust connectivity. However, a fully meshed core layer design is very expensive for a corporation with many divisions.
The number of links in a full mesh is n(n 1)/2, where n is the number of routers. As the number of routers increases, the cost of full-mesh connectivity might become prohibitive.
Hub-and-Spoke Core Layer
As a network grows, fully meshing all the core routers can become difficult. At that point, consolidation into geographically separate data centers might be appropriate. For example, in many companies, data travels to a centralized headquarters where the corporate databases and network services reside. To reflect this corporate centralization, the core layer hub-and-spoke configuration establishes a focal point for the data flow at a key site. The hub-and-spoke design, illustrated in Figure 1-4, supports the traffic flow through the corporation.
Figure 1-4 In a Hub-and-Spoke Core, Each Division Is Connected Only to the Headquarters
A partial-mesh design is also possible, including some nodes connected in a full mesh and some connected in hub-and-spoke fashion.
Access and Distribution Layers
Remote sites are points of entry to the network for end users and customers. Within the network, remote sites gain access to network services through the access layer. The distribution layer consolidates the services and devices that the access layer needs to process the activity that is generated by the remote sites. Figure 1-5 illustrates this process.
Figure 1-5 The Distribution Layer Consolidates Access Layer Connectivity
Frame Relay, shown in Figure 1-5, is a WAN access protocol commonly used to interconnect geographically dispersed sites.
Services should be placed in the distribution layer when there is no benefit to having duplicated services at the remote sites. These services may include Dynamic Host Configuration Protocol (DHCP), Domain Name System (DNS), human resources, and accounting servers. One or more distribution layers can connect to each entry point at the core layer.
You can fully mesh connectivity between remote sites at the access layer. However, using a hub-and-spoke configuration by connecting remote sites to at least two distribution layer devices provides redundancy and is relatively easy to administer.
Benefits of a Good Network Design
An effective network design accommodates unexpected growth and quick changes in the corporate environment. The network design can be adapted to accommodate mergers with other companies, corporate restructuring, and downsizing with minimal impact on the portions of the network that do not change.
The following are characteristics of a good IP addressing plan implemented in a well-designed network:
ScalabilityA well-designed network allows for significant increases in the number of supported sites.
PredictabilityA well-designed network exhibits predictable behavior and performance.
FlexibilityA well-designed network minimizes the impact of additions, changes, or removals within the network.
These characteristics are described further in the following sections.
Scalability of a Good Network Design
Private addresses are reserved IPv4 addresses to be used only internally within a company's network. These private addresses are not to be used on the Internet, so they must be mapped to a company's external registered address when you send anything to a recipient on the Internet.
Key Point: IPv4 Private Addresses
RFC 1918, Address Allocation for Private Internets (available at http://www.cse.ohio-state.edu/cgi-bin/rfc/rfc1918.html), has set aside the following IPv4 address space for private use:
Class A network10.0.0.0 to 10.255.255.255
Class B network172.16.0.0 to 172.31.255.255
Class C network192.168.0.0 to 192.168.255.255
The examples in this book use only private addressing.
The current proliferation of corporate mergers emphasizes the design issues inherent in private IPv4 addressing. For example, if two companies merge, and both use network 10.0.0.0 addresses, there will likely be some overlapping addressing space.
A scalable network that integrates private addressing with a good IP addressing plan minimizes the impact of additions or reorganizations of divisions to a network. A scalable network allows companies that merge to connect at the core layer. Implementing NAT on routers allows the network administrator to translate overlapping network numbers to an unused address space as a temporary solution. Then, the overlapping network numbers can be changed on the devices and/or on the DHCP server in the network.
Good network design also facilitates the process of adding routers to an existing network. For example, in Figure 1-6, two companies have merged. Both companies were using network 10.0.0.0 for addressing. One correct way to merge the two networks would be as follows:
Attach routers P and Q in the new domain to the other routers in the core layer of the network (routers A, D, J, K, X, and Y).
Configure NAT on routers P and Q to change the IP address space of the new company from network 10.0.0.0 to network 172.16.0.0.
Change the DHCP servers to reflect the newly assigned address space in the new network.
Remove NAT from routers P and Q.
Figure 1-6 A Good IP Addressing Design Minimizes the Impact of Merging Networks
Predictability of a Good Network Design
The behavior of a scalable network is predictable.
Packets load-balance across the internetwork if equal-cost paths exist between any two routers in the internetwork. When a circuit or router fails, an alternative equal-cost path to the destination that exists in every routing table can be used, without any recalculation. This alternative path reduces convergence times and route recalculation to typically less than 1 second after the failed circuit or router is discovered.
Depending on the routing protocol used, the equal cost is determined based on hop count and/or bandwidth. For example, if the Routing Information Protocol (RIP) is used in the network shown in Figure 1-6, the routing table for router C will have two best paths to X: three hops through B and three hops through E. Routers B and E each have two best paths to the networks behind router X: Both have two hops through either routers A or D. If router D fails, routers B and E do not need to discover alternative routes because the preferred route exists in the routing table. Thus, if router D fails, the routes to X in router C's routing table do not change.
If a routing protocol that uses bandwidth in its calculation is used (for example, Interior Gateway Routing Protocol [IGRP]), the bandwidth should be configured equally on all interfaces within a layer at each site. For example, in Figure 1-6, routers B and E are consolidation points for the access layer routers (G, C, and F in the example). Routers C, B, and E all have the same bandwidth configured on the links that connect them so that load balancing can be used.
The paths between routers B and E and routers A and D need larger-bandwidth pipes to consolidate the traffic between corporate divisions. Because routers A and D consolidate multiple distribution points for a division, the connections for these routers to other divisions in the company need the largest bandwidth.
The result is a predictable traffic pattern. This level of network behavior predictability is a direct benefit of a scalable network design.
Flexibility of a Good Network Design
A scalable network also needs to be flexible. For example, corporate reorganizations can have minimal impact on the rest of the network when implemented in a scalable network. In the sample network shown in Figure 1-6, assume that Frame Relay is used at the remote sites and that Division Beta is sold and merged with another company, except for remote site H, which becomes part of Division Alpha.
The network administrator in this sample network could accommodate the corporate reorganization with the following process:
Install two additional virtual circuits from router H to routers B and E.
Following a successful installation, remove the virtual circuits to routers M and L.
Perform NAT on the router H interfaces to routers B and E to use the address space of Division Alpha.
Remove the circuits from routers J and K to the other core routers A, D, X, and Y (and P and Q if they are connected).
Change the user addresses for router H to the Division Alpha block of addresses.
Benefits of an Optimized IP Addressing Plan
An optimized IP addressing plan uses hierarchical addressing.
Perhaps the best-known addressing hierarchy is the telephone network. The telephone network uses a hierarchical numbering scheme that includes country codes, area codes, and local exchange numbers. For example, if you are in San Jose, California, and you call someone else in San Jose, you dial the San Jose local exchange number, 528, and the person's four digit number. Upon seeing the number 528, the central office recognizes that the destination telephone is within its area, so it looks up the four digit number and transfers the call.
In many places in North America now, the area code must also be dialed for local calls. This is because of changes in the use of specific digits for area codes and local exchange numbers. The telephone network is suffering from address exhaustion, just like the IP network. Changes in how telephone numbers are used is one solution being implemented to solve this problem.
In another example (see Figure 1-7), to call Aunt Judy in Alexandria, Virginia, from San Jose, you dial 1, and then the area code 703, and then the Alexandria prefix 555, and then Aunt Judy's local number, 1212. The central office first sees the number 1, indicating a remote call, and then looks up the number 703. The central office immediately routes the call to a central office in Alexandria. The San Jose central office does not know exactly where 555-1212 is in Alexandria, nor does it have to. It needs to know only the area codes, which summarize the local telephone numbers within an area.
Figure 1-7 The Telephone Network Uses an Addressing Hierarchy
As you might have noticed, the telephone number used in this example is the number for international directory assistance; it is used for illustration purposes to ensure that Aunt Judy's personal number is not published.
If there were no hierarchical structure, every central office would need to have every telephone number worldwide in its locator table. Instead, the central offices have summary numbers, such as area codes and country codes. A summary number (address) represents a group of numbers. For example, an area code such as 408 is a summary number for the San Jose area. In other words, if you dial 1-408 from anywhere in the U.S. or Canada, followed by a seven-digit telephone number, the central office routes the call to a San Jose central office. Similarly, a routed network can employ a hierarchical addressing scheme to take advantage of those same benefits.
Here are some of the benefits of hierarchical addressing:
Reduced number of routing table entriesWhether it is with your Internet routers or your internal routers, you should try to keep your routing tables as small as possible by using route summarization. Route summarization is a way of having a single IP address represent a collection of IP addresses; this is most easily accomplished when you employ a hierarchical addressing plan. By summarizing routes, you can keep your routing table entries (on the routers that receive the summarized routes) manageable, which offers the following benefits:
More efficient routing
A reduced number of CPU cycles when recalculating a routing table or sorting through the routing table entries to find a match
Reduced router memory requirements
Reduced bandwidth required to send the fewer, smaller routing updates
Faster convergence after a change in the network
Increased network stability
Efficient allocation of addressesHierarchical addressing lets you take advantage of all possible addresses because you group them contiguously. With random address assignment, you might end up wasting groups of addresses because of addressing conflicts. For example, classful routing protocols (discussed in the later section "Implementing VLSM in a Scalable Network") automatically create summary routes at a network boundary. Therefore, these protocols do not support discontiguous addressing (as you will see in Chapter 2, "Routing Principles"), so some addresses would be unusable if not assigned contiguously.
Within the context of hierarchical addressing, the IP addressing plan must include provisions for summarization at key points. Summarization (also called aggregation or information hiding) is not a new concept. When a router announces a route to a given network, the route is a summarization of the addresses in the routing table for all the host devices and individual addresses that reside on that network.
Summarization helps reduce routing-table size and helps localize topology changes. This promotes network stability because a reduced routing-table size means that less bandwidth, memory, and CPU cycles are required to calculate the best path selection. Because summarization limits the propagation of detailed routes, it also reduces the impact to the network when these detailed routes fail.
Scalable Network Addressing Example
The network illustrated in Figure 1-8 shows an example of scalable addressing. In this example, a U.S. national drugstore chain plans to have a retail outlet in every city in the country with a population greater than 10,000. Each of the 50 states has up to 100 stores, with two Ethernet LANs in each store:
One LAN is used to track customer prescriptions and pharmacy inventory and reorder stock.
The second LAN is used to stock the rest of the store and connect the cash registers to a corporate-wide, instantaneous point-of-sale evaluation tool.
Figure 1-8 Scalable Addressing Allows Summarization
The total number of Ethernet LAN networks is 50 states * 100 stores per state * 2 LANs per store = 10,000. (An equal number of serial links interconnect these stores.)
Using a scalable design and creating 51 divisions (one for each state and one for the backbone interconnecting the divisions), the corporation can assign each division a block of IP addresses 10.x.0.0 /16. Each LAN is assigned a /24 subnet of network 10.0.0.0, and each division has 200 such subnets (two for each of the 100 stores). The network will have 10,000 subnets; without summarization, each of the 5000 routers will have all these networks in their routing tables.
If each division router summarizes its block of networks 10.x.0.0 /16 at the entry point to the core network, any router in a division has only the 200 /24 subnets within that division, plus the 49 10.x.0.0 /16 summarizations that represent the other divisions, in its routing table. This results in a total of 249 networks in each IP routing table.
Nonscalable Network Addressing
In contrast to the previous example, if a hierarchical addressing plan is not used, summarization is not possible, as is the case in Figure 1-9. Problems can occur in this network related to the frequency and size of routing table updates and how topology changes are processed in summarized and unsummarized networks. These problems are described next.
Figure 1-9 Nonscalable Addressing Results in Large Routing Tables
Routing protocols such as RIP and IGRP, which send a periodic update every 30 and 90 seconds, respectively, use valuable bandwidth to maintain a table without summarization. A single RIP update packet is limited to carrying 25 routes; therefore, 10,000 routes means that RIP on every router must create and send 400 packets every 30 seconds. With summarized routes, the 249 routes means that only 10 packets need to be sent every 30 seconds.
Unsummarized Internetwork Topology Changes
A routing table with 10,000 entries constantly changes. To illustrate this constant change, consider the sample network with a router at each of 5000 different sites. A power outage occurs at site A, a backhoe digs a trench at site B, a newly-hired system administrator begins work at site C, a Cisco IOS software upgrade is in progress at site D, and a newly-added router is being installed at site E.
Every time a route changes, all the routing tables must be updated. For example, when using a routing protocol such as Open Shortest Path First (OSPF), an upgrade or topology change on the internetwork causes a shortest path first (SPF) calculation. The SPF calculations are large because each router needs to calculate all known pathways to each of the 10,000 networks. Each change a router receives requires time and CPU resources to process.
Summarized Network Topology Changes
In contrast to an unsummarized network, a summarized network responds efficiently to network changes. For example, in the sample drugstore network with 200 routes for each division, the routers within the division see all the subnets for that division. When a change occurs on one of the 200 routes in the division, all other routers in the division recalculate to reflect the topology change of those affected networks. However, the core router of that division passes a summarized /16 route and suppresses the /24 networks from advertisement to the core routers of other divisions. The summarized route is announced as long as any portion of the summarized block can be reached from that core router. The more-specific routes are suppressed so that changes from this division are not propagated to other divisions.
In this scenario, each router has only 200 /24 networks, compared to the 10,000 /24 networks in an unsummarized environment. Obviously, the amount of CPU resources, memory, and bandwidth required for the 200 networks is less than the 10,000 networks. With summarization, each division hides more-specific information from the other divisions and passes only the summarized route that represents that overall division.