A single-path routing infrastructure is one in which a unicast IP packet can take only a single route between any source node and any destination node. Although single-path routing simplifies routing tables and packet flow paths, single-path internetworks are not fault tolerant. On an internetwork in which all routing information is manually configured, a router cannot detect a failure, and if a failure occurs, networks beyond the failure are unreachable until the failure is resolved. By contrast, a router running a dynamic routing protocol can detect a failure, but, because only a single path exists, networks beyond the failure are also unreachable for a dynamic router for the duration of the failure. A downed link or router must be brought back up before packets can again be delivered successfully to all locations on the internetwork. Single-path routing is thus not feasible for a large internetwork that depends on consistent network availability.

By contrast, a multipath routing infrastructure is one in which a unicast IP packet can take one of several paths between any source node and any destination node. A multipath IP internetwork that uses dynamic routers is fault tolerant because more than one path exists, which means that routers can not only detect a failure but can also route packets to networks beyond the downed link or router by using an alternative path.

Typically, multipath internetworks provide more reliable network availability than single-path internetworks but are more complex to configure. In addition, although multipath networks that use distance vector–based routing protocols provide better availability than single-path networks, they can experience routing loops if configuration errors occur or during convergence (the process by which routers update routing tables after a change in network topology and reestablish a stable state). A multipath network that uses a link state routing protocol, such as OSPF, provides more efficient convergence and thus better stability than a distance vector routing protocol, such as RIP, but requires more effort to plan and deploy.

For more information about static and dynamic routing, see How Static and Dynamic Routing Work later in this document.

An internetwork that uses a flat routing infrastructure consists of multiple network segments connected to each other without a parent-child structure. All routers are peers, and each network segment is stored as a separate route in the routing table. The network IDs have no network or subnet structure and routes cannot be summarized, which means it is not possible to group a set of network segments as a single entry in the routing table. Therefore, although flat routing can work well for small- or medium-sized IP internetworks, it is not well-suited to large internetworks because it requires that its routers store an ever-increasing number of routes in their routing tables.

By contrast, an IP internetwork that uses a hierarchical routing infrastructure consists of multiple network segments sharing the same network ID prefix grouped into a hierarchy that consists of a network, subnet, and sub-subnet structure. Each group of network segments in the hierarchy can be represented as a single network ID through route summarization and can therefore be represented in the IP routing table as a single route. The routing table entry for the highest level (the network) is also the route used for regions of the network and for individual subnets.

In a hierarchical routing infrastructure, a collection of contiguous networks connected by routers that share routing information is referred to as a routing region (also known as a routing domain or routing area). One part of the internetwork that is divided into routing regions can be connected by a common routing region, called a backbone, to another part of the internetwork that is also divided into routing regions. Routers within each routing region perform intradomain routing. Routers connected to the backbone perform interdomain routing. Only a relatively few routes are required on the backbone of a well-planned hierarchical internetwork.

Some IP internetworks use both flat and hierarchical routing. For example, the Internet, originally designed as a flat routing infrastructure until the Internet community helped pioneer the introduction of classless addressing and summarized routing, now incorporates both flat and hierarchical routing. One of the factors that results in the storing of large numbers of routes in the routing tables of Internet backbone routers is the continued inclusion of flat routing within the Internet. These large Internet backbone routing tables impede the speed at which Internet traffic can be forwarded and is one of several factors driving the development of IPv6. The hierarchical routing infrastructure and new address types provided by IPv6 are designed to require relatively few routing entries in the routing tables of Internet backbone routers.

For more information about route summarization on a hierarchical IPv4 internetwork, see “CIDR Blocks and Hierarchical Routing Enabled by Route Summarization” in How IP Addressing and IP Routing Interact later in this document.

Very large IP internetworks are divided into separate entities called autonomous systems. An autonomous system (AS) is a portion of the internetwork — either a single network or a collection of networks — under a common administrative authority, such as an enterprise, a business division, or other organization, that shares a common routing protocol. The autonomous system is described in RFC 1930, “Guidelines for creation, selection, and registration of an Autonomous System (AS)” in the IETF RFC Database.

An autonomous system is sometimes said to be defined by the use of the shared routing protocol that it uses. For example, a contiguous portion of an IP internetwork that uses OSPF to exchange routing information among a group of routers is under OSPF administrative authority and is, therefore, an OSPF autonomous system.

In a large organization, one autonomous system might be subdivided into multiple routing regions (routing domains or areas) that define a hierarchy within that autonomous system. Regions within an autonomous system can use summarized routes for traffic between the regions to decrease the number of routes exchanged and stored in routing tables.

Types of autonomous system

Autonomous system numbers

An IP source host determines the next-hop IP address of a neighboring router by using one of the following methods:

IP address currently in the host routing table

A routing table on the host provides the IP address of a router that can forward the packet towards its destination. For more information about routing tables, see How the IPv4 Routing Table Works later in this document.

ICMP dynamic updates of the host routing table

If a sending host forwards a packet to a neighboring router but another neighboring router is closer to the destination, the router that initially receives the packet can send an ICMP Redirect message to the sending host, informing it of the better router to use as the next-hop for the destination. When the sending host receives the Redirect message, it adds a host route for the destination to the local routing table. TCP/IP for Windows Server 2003 supports the dynamic update of the IP routing table based on the receipt of ICMP Redirect messages.

Default route

A default route, which summarizes destinations that are not located on the local subnet, is the route on which IP packets are sent when no other route to the destination network is found in the routing table. To simplify the configuration of IP nodes and to reduce the overhead that exists if each host stored routes for all network IDs on the internetwork, a source host is configured with a single default route. In IPv4, the default route is {Network ID: 0.0.0.0, Subnet Mask: 0.0.0.0}, also expressed as 0.0.0.0/0. For more information about the default route, see How the IPv4 Routing Table Works later in this document.

Eavesdropping

IP hosts can listen to routing protocol traffic if the capability known as eavesdropping or wiretapping is enabled. Eavesdropping hosts maintain the same routing information as do routers. An example of eavesdropping is silent RIP, which is the ability of an IP host to listen to RIP routing traffic exchanged by RIP routers and to update the routing table on the host. A computer that uses silent RIP processes RIP announcements but does not announce its own routes. You can use silent RIP on a non-router host (such as a workstation or a server running Routing and Remote Access that has only one interface and is therefore not a router) to produce a routing table that contains as much detail as the routing table on a RIP router. By including detailed routes in its routing table, a silent RIP host can perform routing more efficiently.

For example, you can configure computers running Windows XP as silent RIP hosts in an environment that includes RIP routers by installing the RIP Listener networking component. Instead of a user or administrator manually adding routes to the routing table, the computer learns other routes on the network by listening to RIP messages and then adds advertised routes in those messages to its routing table. You do not need to configure a silent RIP host with a default gateway (IP address of a directly reachable IP router).

The eavesdropping provided by silent RIP can be useful in some remote access environments, such as when a remote access computer connects to a corporate network over a dial-up connection. The additional routes added by RIP Listener can improve network access and performance over the dial-up connection by reducing the number of packets that are sent to the incorrect router. Windows Server 2003, Microsoft® Windows® XP, Windows® 2000, Microsoft Windows NT® Workstation version 4.0 Service Pack 4 (SP4) and later, and Windows NT Server version 3.51 Service Pack 2 (SP2) and later support Silent RIP.

Some routable protocols, including IP, can enable the source host to do more than simply determine the first hop. In this scenario, IP enables the source host to determine the path between the source host and the destination, and then includes the list of routers in the network layer header of the packet. The network layer header is used by each router to forward the packet along the indicated path. This process is known generically as source routing.

Source routing is not typically implemented on IP internetworks because in source routing the path must be known. Typically, instead of using source routing, IP routing decisions are made by source hosts and IP routers based on their respective local routing tables. However, in network testing and debugging situations, you might sometimes want to specify an exact route that overrides the route as determined by routing tables as the packet travels towards its destination. Specifying an exact route through the IP internetwork is known as IP source routing.

In IP source routing, the source host specifies the entire route through successive IP routers between the source and destination. Each IP router addresses the IP packet to the next router by using the Destination IP Address field in the IP header.

IP supports two types of source routing:

• Loose source routing. The IP address of the next router can be one or more routers away (multiple hops).
  
  
• Strict source routing. The next router must be a neighboring router (single hop).
  
  

Note
Token Ring source routing is a MAC–sublayer routing scheme; it is not related to the internetwork-based source routing discussed here.

Of the five standard address classes, Class A, B, and C addresses, collectively known as IPv4 unicast addresses, are assigned to specific devices on an IPv4 network. Class D addresses, known as multicast addresses, are used for IP multicasting, simultaneously sending a message from one source to more than one network destination (for more information about multicasting, see IPv4 Multicasting Technical Reference). Class E addresses are reserved for experimental purposes.

To be able to configure fixed-size subnetting on your network (as well as variable length subnetting or route summarization, described later), you must first understand the default formats of the unicast IPv4 addresses. All 32-bit IPv4 addresses contain four 8-bit bytes called octets separated for readability by spaces (in binary notation) or represented as four decimal numbers separated by dots (in dotted decimal notation). If all or some of the actual numbers are not known, all or part of the IP address in dotted decimal notation is typically represented by w.x.y.z. The following table shows a single IP address in both decimal and binary notation.

IP Address in Decimal and Binary Notation

Network ID bits and host ID bits

Unicast IPv4 address ranges

The following table shows the address ranges and the number of possible networks and hosts for each type of IPv4 unicast address class. Keep in mind that a network ID is not an IP address, even though both are expressed by using dotted decimal notation.

Classful Unicast IPv4 Address Classes

Class	Leading Bit Pattern	First Network ID in Range	Last Network ID in Range	Available Networks	Hosts per Network
A	0	1.0.0.0	126.0.0.0	126	16,777,214
B	10	128.0.0.0	191.255.0.0	16,384	65,534
C	110	192.0.0.0	223.255.255.0	2,097,152	254

Note
The network ID 127.0.0.0 is reserved for loopback addresses, which are used for testing TCP/IP operation.

The range in the first octet is used to identify the address class. For example, an address in the format 10.x.y.z is immediately recognizable as a class A address because 10 falls within the range 1–126. Similarly, an address in the format 131.107.y.z is identifiable as a class B address, and an address in the format 206.73.118.z is a class C address.

Classful subnet masks

Without some means of subdividing networks, all available IPv4 addresses would have been depleted long ago. Classful subnetting, dividing a network into subnets of equal size rather than simply using the default classful formats, is the initial strategy that was defined in RFC 950 to handle this problem and is still often used in smaller organizations.

The RIP v1 routing protocol, which does not advertise a subnet mask with the route, can reliably route only fixed-size subnets. RIP v1 advertises only the class-based network ID and therefore cannot perform routing in an environment that includes variable-size subnets.

Using host ID bits to configure multiple subnets

Using succinct network prefix notation to indicate subnetting

Classless routing requires the following addressing mechanisms:

Subnet mask explicitly advertised by routing protocol

{IP Address, Subnet Mask} pair used to derive network ID

{Network ID, Subnet Mask} pair used to express a summary route

Longest match routing

As described earlier, standard subnetting creates subnets of equal size in a networking environment that supports only classful routing protocols. Subnets of equal size can be inefficient if each subnet does not need to support the same number of hosts. Therefore, a more useful method of subnetting, known as variable length subnetting, can be configured by using variable length subnet masks (VLSMs). Classless routing protocols support variable length subnetting, which is defined in RFC 1878.

Using VLSMs, you can perform subnetting on the same address space (network ID) multiple times to produce subnets of different sizes on the same network. All subnetted network IDs are unique and can be distinguished from each other by their corresponding subnet mask.

For example, by starting with a classful class B network ID of 131.107.0.0/16, you can configure one subnet that supports 1,022 hosts, a second that supports 510 hosts, and a third that supports only two hosts (for a point-to-point WAN link between two routers). The following table shows how.

Example Using Three VLSMs to Create Three Various-Sized Subnets from 131.107.0.0

In this example, creating three subnets, each supporting a variable number of hosts, is the hypothetical goal. However, this simple example does not exhaust the possibilities. Instead of simply creating three subnets with 1,022, 510, and 2 hosts, respectively, you could create a more complex result by taking advantage of the number of possible subnets as well as the number of possible hosts. Because 131.107.0.0/16 is a class B network ID, the number of possible subnets is determined by the number of bits set to 1 in the third and fourth octets:

• 22-bit prefix length (Subnet 1): 64 (26) possible subnets, each supporting up to 1,022 hosts.
  
  
• 23-bit prefix length (Subnet 2): 128 (27) possible subnets, each supporting up to 510 hosts.
  
  
• 30-bit prefix length (Subnet 3): 16,384 (214) possible subnets, each supporting up to 2 hosts.
  
  

When using VLSM, make sure that you do not overlap blocks of addresses. A good technique is to start with equal-size subnets and then subdivide them.

If variable-length subnets are also contiguous, the routes for all of the subnets can be summarized: a router can be configured to advertise the set of subnets to another router by advertising only the original network ID. Route summarization is described in the next section.

This section describes how classless routing uses a summary route to represent a range of network IDs (also known as a CIDR block) and how a multi-level hierarchy uses summary routes at each level of the hierarchy to make routing more efficient.

How a summary route represents a range of network IDs

How a multi-level hierarchy using summary routes streamlines routing

Hierarchical routing requires a method of address allocation that can reflect a hierarchical network topology and that can reduce the size of routing tables. Hierarchical routing accomplishes both requirements by using a single route to advertise a set of routes at each level of the hierarchy.

ISPs, organizations that subscribe to ISPs, and subnets within organizations are some of the typical levels in the routing hierarchy of the public Internet. An element lower in the hierarchy reports summary routing information to the level above it. For example, the allocation of large blocks of the public IPv4 address space to network service providers (such as ISPs), which then allocate subsets of these blocks to smaller organizations (such as organizations that subscribe to an ISP) is a practice recommended in RFCs 1518 and 1519.

Thus, a router within an organization does not need to advertise routing information about every subnet in that organization. Instead, the router can be configured to maintain routing information on a per region basis, in which each routing region is summarized with a single route except for the regions to which the router is connected. Likewise, an ISP router does not need to advertise routing information about individual subnets within each of the organizations that make up the set of its subscribers. Instead, the router can maintain routing information on a per subscriber basis.

Classless routing protocols, which recognize a summarized route represented by a {Network ID, Subnet Mask} pair that represents a set of reachable IP addresses, are designed to handle routing in a hierarchical infrastructure. By expressing a range of network IDs as a single route in the routing table at each level of the hierarchy, route summarization decreases the number of routes on routers. With fewer routes to check, routers can forward packets faster. One major use of route summarization is to reduce the burden for routers on the Internet backbone.

The example shown in the following figure depicts a simple hierarchical routing structure divided into routing regions based on the private network ID 10.0.0.0/8. Route summarization is used at the border of each region. Each arrow is labeled with the summarized route that is advertised outside its region by the router or routers at the border of that region. (This example figure is used by permission from the authors of "Microsoft Windows Server 2003 TCP/IP Protocols and Services Technical Reference" by Joseph Davies and Thomas Lee, Redmond, WA, Microsoft Press, 2003.)

The structure shown in this example illustrates both the advantage of hierarchical routing and its cost:

• Advantage. The efficiency provided by hierarchical routing reduces the size of routing tables, reduces the amount of communication required between routers in order for them to update their routing tables, and therefore reduces network traffic. With smaller routing tables, routers forward packets faster.
  
  
• Cost. The initial planning required to establish such a structure takes time. In a large organization, this initial investment in time is well worth the effort because of the long-term efficiency gained.
  
  

An administrator or user can view the current state of the IP routing table on an IP node at any time. For example, consider a node configured as follows:

• IP address: 157.54.27.90
  
  
• Subnet mask: 255.255.240.0
  
  
• Default gateway: 157.54.16.1
  
  

In network prefix length notation, the IP address and subnet mask pair {157.54.27.90, 255.255.240.0} for this node is expressed as 157.54.27.90/20. Typing route print at a command prompt on this node produces the output shown in the following table.

Example Routing Table

Network Destination      Netmask        Gateway      Interface   Metric
        0.0.0.0           0.0.0.0    157.54.16.1   157.54.27.90      20
      127.0.0.0         255.0.0.0      127.0.0.1      127.0.0.1       1
    157.54.16.0     255.255.240.0   157.54.27.90   157.54.27.90      20
   157.54.27.90   255.255.255.255      127.0.0.1      127.0.0.1      20
 157.54.255.255   255.255.255.255   157.54.27.90   157.54.27.90      20
      224.0.0.0         240.0.0.0   157.54.27.90   157.54.27.90      20
255.255.255.255   255.255.255.255   157.54.27.90   157.54.27.90       1

This routing table, displayed on a computer running Windows Server 2003, differs from the routing table on a computer running Windows 2000 Server in the following ways:

• New netmask for the multicast address route. In the routing table on a server running Windows 2000 Server, the Network Destination field and the Netmask field for the route used for IPv4 multicast addresses (defined as those in the range from 224.0.0.0 through 239.255.255.255) are both 224.0.0.0. However, the subnet mask used for IPv4 multicast addresses is 240.0.0.0. Therefore, the new Netmask of 240.0.0.0 that is used in the Windows Server 2003 routing table is more appropriate than the 224.0.0.0 Netmask used in Windows 2000, based on the class D definition of IPv4 multicast addresses.
  
  
• Automatic metric feature. On a server running Windows 2000 Server, the default routing metric for each route (each row) in the routing table is set to 1. On a computer running Window Server 2003 or Windows XP, the routing metric for each route is set to a value determined by the link speed of the interface. (This automatic metric feature does not apply to the default route metric when the default gateway is manually configured with a metric.)
  
  

In the output shown in “Example Routing Table,” the routing metric that is set to 20 for several rows is the metric that Windows Server 2003 or Windows XP TCP/IP uses for a 100-Mbps Ethernet interface, for an 802.11g adapter, or for an 802.11a adapter. The following table lists the criteria that Windows Server 2003 or Windows XP uses to assign metrics for routes that are bound to network interfaces of varying speeds.

Metrics Assigned by the Automatic Metric Feature

The automatic metric feature is enabled by default through the Automatic metric check box on the IP Settings tab on Advanced TCP/IP Settings of the TCP/IP protocol. For DHCP-assigned default gateways, you can override the default behavior of automatically calculating a metric for the default route based on adapter speed by using the Microsoft-specific DHCP option called Default Router Metric Base.

Many router implementations, including the Windows Server 2003 Routing and Remote Access service, use both a routing table and a forwarding table:

• Routing table. Stores all routes from all possible sources. The command netsh ro ip sh rtmr (that is, netsh routing ip show rtmroutes) shows the routing table.
  
  
• Forwarding table. Used by IP to forward packets. The commands route print or netsh ro ip sh rtmd (that is, netsh routing ip show rtmdestinations) show the IP forwarding table.
  
  

The Routing and Remote Access service on a Windows Server 2003 router maintains the IP routing table by using a component called Route Table Manager (abbreviated as rtm in rtmroutes and rtmdestinations in the commands shown earlier). Route Table Manager updates the IP forwarding table (which is contained within the TCP/IP protocol) based on incoming route information from multiple sources.

The contents of the routing table do not necessarily match the contents of the forwarding table. However, in this introductory discussion, the routing table and the forwarding table are treated as if they are identical.

On an IPv4 internetwork, all routing decisions made by a host or a router are based on information stored in the IP routing table on that node. Because the routing table is a local database that physically resides in the remote access memory (RAM) of the node making the routing decision, the IP routing table is rebuilt whenever the node restarts. Thus, there is no single, unified view of the internetwork that is gathered by a server and downloaded to each host and router — all nodes do not have the same view of the internetwork, and all traffic does not flow along predictable pathways.

Each router in a path between a source and a destination makes a local routing decision based on its local routing table. Therefore, the path that packets take from the source to the destination might not be the same as the path taken by response packets from the destination back to the source.

If the information in the local routing table of a node is incorrect due to configuration errors or changing network conditions, routing problems can result. Troubleshooting routing problems might involve the analysis of the routing tables of the hosts (source and destination) and analysis of all of the routers that forward packets between them.

The following table describes the types of routes stored in an IP routing table. For a description of each route in an actual routing table, see the table “Routes in an IP Routing Table” later in this section.

Types of Routes Stored in an IP Routing Table

Configuring a default gateway creates a default route in the IP routing table. For IP nodes, the default gateway (also called a default router) is a neighboring IP router that forwards unicast traffic for the node by providing a next-hop IP address (the Gateway column in the routing table) and interface (the Interface column in the routing table) for all destinations that are not located on the local subnet. The default gateway address is the IP address of a directly reachable IP router.

The default gateway on a computer running Windows Server 2003 or Windows XP is configured by using one of the following methods:

• TCP/IP properties General tab. If you want the node to obtain its IP address configuration by using manual configuration, the default gateway is the IP address located in the Default gateway field on the General tab on the TCP/IP properties page. Multiple default gateways are configurable by adding them on the IP Settings tab on the Advanced TCP/IP Settings page of TCP/IP properties.
  
  
• TCP/IP properties Alternate Configuration tab. If you want the node to obtain its IP address configuration by using the user-configured alternate configuration option, the default gateway is the IP address located in the Default gateway field on the Alternate Configuration tab on the TCP/IP properties page. You can specify only a single default gateway. The TCP/IP alternate configuration feature lets a computer function in two or more networks — one configured with static IP addresses and another configured with DHCP — without reconfiguration of network adapter parameters. When no DHCP server is available, the connection uses the configuration specified on the Alternate Configuration tab.
  
  
• DHCP. If you want the node to obtain its IP address configuration by using DHCP, the default gateway is the value of the first IP address in the Router DHCP option. The Router DHCP option specifies an ordered list of one or more default gateways. If you are using a Windows Server 2003–based DHCP server, the Router option is located in the Details pane in the DHCP snap-in console tree under ServerName\ScopeName\Scope Options.
  
  
• ICMP Router Discovery. If you want the node to automatically discover the best default gateway router available on a subnet, you can configure nodes on that subnet to use ICMP router discovery. ICMP Router Solicitation messages and Router Advertisement messages exchanged between routers and hosts enable hosts to dynamically discover which local routers are available, which routers are down, and which router is currently the best default gateway to use on a subnet. By using ICMP router discovery, a host can automatically switch to another default gateway if its current default gateway becomes unavailable. For information about support for ICMP router discovery in the Windows Server 2003 Routing and Remote Access service, see “How ICMP Router Discovery Works” in How Unicast IPv4 Routing Protocols and Services Work.
  
  

Note
If the host obtains its IP address configuration by using Automatic Private IP Addressing (APIPA), no default gateway is configured and no default route is created in the routing table. APIPA is useful only for a single subnet.

As with any database, understanding the IP routing table requires understanding the relationship between the records (rows) and the fields (columns) that make up the IP routing table. This relationship is best understood by using an example. The following table shows the same information displayed in “Example Routing Table” earlier in this section. It labels each row with the route type for that entry.

Example Information Stored in an IP Routing Table

Routing table columns

Routing table rows

The following table summarizes how IP makes use of the IP routing table.

Overview of How IP Uses Routing Table Rows and Columns

The next section describes in detail the process by which IP determines which routing decision to make.

IP determines the best route to a destination by comparing the destination IP address in the packet to each route currently present in the routing table. The detailed process is as follows:

1. Determine whether the packet’s destination IP address matches one or more routes. For each entry in the routing table, IP on the node performs a bitwise logical AND operation between the destination IP address of the packet and the subnet mask listed in the Netmask column. IP compares the resulting value to the value in the Network Destination column:
   
   
   • Match. If one or more matching routes are found, IP compiles a list of the matching routes. IP determines if a route produces a match as follows:
     
     Host route. All 32 bits match the destination IP address. The route is to a single destination address.
     
     Route representing the network ID for a local or remote subnet. All of the bits in the network ID match the destination IP address. This route is either a route to a destination on the local subnet, or it is a route to a destination on a non-local network through a router.
     
     Route representing a summarized network ID. All of the bits in the summarized network ID match the destination IP address. The route is to a destination on the set of subnets summarized by this summary route.
     
     Default route. All destination IP addresses match the default route. This is the route used when there is no more specific match.
     
     
   • No match. If no route is found (including no default route), IP indicates an error condition. A “no match” condition cannot occur if a default route exists. If the node is a host, an IP routing error is sent internally to an upper layer protocol (such as TCP, UDP, or ICMP). If the node is a router, an ICMP Destination Unreachable-Host Unreachable message is sent to the source host.
     
     
   For examples illustrating the AND operation between a destination IP address and the value in the Netmask column, see the Netmask row in the table “Columns for Each Routing Table Entry” in “Structure of the IPv4 Routing Table” earlier in this section.
   
   
2. Determine the single route to use to send or forward the packet. The result of the route determination process is the choice of a single route in the routing table:
   
   
   • If one and only one route has the longest match — the longest match is the route with the highest number of bits set to 1 in the Netmask column (the longest prefix length) — that route is selected.
     
     
   • If multiple longest match routes are found (for example, multiple routes to the same network ID), IP selects the route with the lowest metric.
     
     
   • If multiple longest match routes with the same lowest metric exist, IP selects the route associated with the interface that is first in the binding order.
     
     

   Note

   If a computer has more than one interface (network adapter), the binding order is the order in which the interfaces are accessed by network services. This order reflects the order in which TCP/IP is bound to each of the interfaces. To change the relative binding order of interfaces on a computer running Windows Server 2003 or Windows XP, open Network Connections, select the Advanced menu, select Advanced Settings, click the interface whose binding order you want to change, and then click the up arrow or the down arrow, as appropriate.

3. Determine the next-hop address and interface. After a route is selected, IP determines from the routing table entry what the next-hop IP address is and which interface (physical network adapter or logical port) to use to forward the packet:
   
   
   • Direct delivery (to the destination node). If the destination is on a subnet directly connected to the host or router, IP delivers the packet to the destination node. In this case, the address in the Gateway column is the same as the address in the Interface column (or the Gateway column is blank), and the next-hop IP address is set to the destination IP address in the IP packet. The interface used is the one specified in the Interface column of the selected route.
     
     Again using the example shown in “Example Routing Table” described earlier, if traffic is sent to 157.54.16.48, the most specific route is the route for the directly attached network (157.54.16.0/20). The next-hop IP address is set to the destination IP address (157.54.16.48) and the interface used is the adapter card that has been assigned the IP address 157.54.27.90 (the interface on the local node).
     
     
   • Indirect delivery (to the next router). If the destination is not on a subnet directly connected to the node, IP delivers the packet to a neighboring router on a directly connected subnet for further routing. In this case, the address in the Gateway column is not the same as the address in the Interface column, and the next-hop IP address is set to the address in the Gateway column. The interface used is the one specified in the Interface column of the selected route.
     
     For example, if traffic is sent to 157.60.0.79, the most specific route is the default route (0.0.0.0/0). The next-hop IP address is set to the gateway address (157.54.16.1) and the interface used is, in this case again, the adapter card that has been assigned the IP address 157.54.27.90 (the interface on the local node).
     
     
4. Hand off the packet to ARP. IP hands the packet, the next-hop IP address, and the next-hop interface to ARP. ARP resolves the next-hop IP address to its MAC address and forwards the packet, as appropriate, to the next hop, which is either the destination or the router that will forward it towards its destination.
   
   

For more information about the IP routing process, including details of the host forwarding process, the router forwarding process, and the destination host receiving process, see “IP Routing from Sending Host to Destination” in "Microsoft Windows Server 2003 TCP/IP Protocols and Services Technical Reference" by Joseph Davies and Thomas Lee, Redmond, WA, Microsoft Press, 2003.

An administrator configuring a static routing entry manually specifies the following information:

• Destination. A destination IP address for this route, which can be a host address, a subnet address, a summarized network address, or the destination for the default route (0.0.0.0).
  
  
• Network mask. The subnet mask for the static route. This number is used in conjunction with the destination IP address. For example, a mask of 255.255.255.255 means that only an exact match of the network mask with the destination can use this route. A mask of 0.0.0.0 means that any destination can use this route. For more information, see “Details of How IP Makes the Routing Decision” earlier in this document.
  
  
• Gateway. The next-hop IP address for this route. For a LAN interface, the gateway address is the IP address of a neighboring router. For a demand-dial interface, no gateway address is configured.
  
  
• Metric. The cost (such as the number of hops to the destination, or a unitless metric, such as the speed or reliability of the link) associated with this route. When multiple routes to the same destination exist, IP selects the route with the lowest metric.
  
  

For more information about how IP handles entries in the routing table, see How the IPv4 Routing Table Works earlier in this section.

Using static routing exclusively is appropriate only for a small internetwork that is rarely reconfigured. However, a medium-size or large internetwork that primarily uses dynamic routing typically also uses some manually configured static routes. For example, you might use a static route with a higher metric as the redundant backup for a dynamically configured route. In addition, you might use dynamic routing for most paths but configure a few static paths where you want to ensure that network traffic follows a particular route, such as configuring routers to force traffic over a specific path to a high-bandwidth link.

Using dynamic routing protocols across temporary dial-up WAN links is not feasible because dynamic routing protocols typically advertise the contents of the routing table at periodic intervals. Such automatic periodic advertisements are appropriate only for permanently connected LAN or WAN links. If a dial-up WAN link is configured to use dynamic routing, it might incur long-distance charges each time one router calls another router to establish a connection and update the routing table. A more cost-effective way to update the routing table over a dial-up WAN link is to configure autostatic updates. RIP supports autostatic updates; OSPF does not.

An autostatic route is a static route that is dynamically obtained through an exchange of RIP messages. An administrator can enable autostatic updates on an interface that is used for a demand-dial connection so that — whenever the demand-dial link is in a connected state — a request for updated routes can be sent over that interface. Autostatic updates are automatic only in the sense that, when the administrator gives the command to update routes, the router sends an update request across the link to the router on the other side of the connection, and then the router automatically adds the requested routes as static routes to its routing table. Although the term autostatic update might seem to imply an automatic operation, autostatic updates are not automatically performed when the demand-dial connection is made but only if the command to perform the update is given.

As an alternative to issuing a command to initiate a manual autostatic update, if you use Routing and Remote Access RIP routers, you can use the Routing and Remote Access snap-in, Netsh commands, and Task Scheduler to schedule autostatic updates:

1. In the Routing and Remote Access snap-in, enable autostatic update mode by navigating to the IP Routing\RIP node in the console tree, and then selecting the Auto-static update mode option on the General tab of the InterfaceName Properties page.
   
   
2. Create a batch file or Netsh script file that contains the following commands:
   
   netsh interface set interface name=DemandDialInterfaceName connect=CONNECTED
   
   netsh routing ip rip update DemandDialInterfaceName
   
   netsh interface set interface name=DemandDialInterfaceName connect=DISCONNECTED
   
   
3. Open Scheduled Tasks in Control Panel, double-click Add Scheduled Task, and then follow the wizard instructions to establish the schedule for the batch file or Netsh script to run.
   
   

A computer running the Windows Server 2003 family can act as a static IPv6 router that forwards IPv6 packets between interfaces based on the contents of the IPv6 routing table. You can configure static routes by using the netsh interface ipv6 add route command.

For information about Windows Server 2003 support for static routing in an IPv6 environment, see “How IPv6 Works” in IPv6 Technical Reference.

An important element of a routing protocol implementation is its ability to sense and recover from internetwork faults. How quickly the routing protocol can enable recovery from failures on the internetwork is determined by the type of failure, how it is sensed, and how the routing information is propagated throughout the internetwork.

When all routers on an internetwork have the correct routing information in their routing tables, the internetwork is said to have converged. Convergence is the process by which routers update routing tables after a change in network topology occurs — the change is replicated to all routers that need to know about it. When convergence is achieved, the internetwork is in a stable state and all routing occurs along optimal paths.

When a link or router fails, the internetwork must reconfigure itself to reflect the new topology. Information in routing tables must be updated. Until the internetwork reconverges, it is in an unstable state in which routing loops and black holes (both described later) can occur. The time it takes for the internetwork to reconverge is known as the convergence time. The convergence time varies based on the routing protocol and the type of failure — downed link or downed router.

For more information about convergence, see “Impact of Routing Loops and Black Holes During Convergence” later in this section, and see “How RIP Works” and “How OSPF Works” in How Unicast IPv4 Routing Protocols and Services Work.

Autonomous systems (described earlier in “IPv4 Routing Infrastructure”) use two types of routing protocols to update routing information:

• The protocols used to distribute routing information between two or more autonomous systems are known as Exterior Gateway Protocols (EGPs).
  
  
• The protocols used to distribute routing information within a single autonomous system are known as Interior Gateway Protocols (IGPs).
  
  

The following figure depicts two autonomous systems using interior and exterior protocols to communicate.

EGPs are inter–autonomous system routing protocols. EGPs define the way that all routes within the autonomous system are advertised outside of the autonomous system. Routers that connect autonomous systems to the Internet backbone use an EGP to advertise routing information to each other. Advertising can include a list of routes in a flat routing infrastructure or a list of summarized routes in a hierarchical routing infrastructure.

EGPs are independent of the IGPs used within an autonomous system and can enable the exchange of routes between autonomous systems that use different IGPs.

The EGPs for IP internetworks include:

• Exterior Gateway Protocol (EGP). An obsolete EGP that was developed to communicate information between autonomous systems on the Internet. Although EGP is the eponym for the generic protocol type for inter-autonomous system routing protocols, it is no longer used on the Internet because it cannot support multipath environments or classless routing. EGP was defined in the now obsolete RFC 904, “Exterior Gateway Protocol.”
  
  
• Border Gateway Protocol (BGP). The EGP that is currently used to communicate information between autonomous systems on the Internet. The Internet is a large IP internetwork divided into several autonomous systems that are connected by the Internet’s core routers, which use BGP for communication among themselves. BGP uses autonomous system numbers (ASNs) to avoid routing loops and to implement policy-based routing on the Internet backbone. Unlike its obsolete predecessor EGP, BGP supports complex multipath networks and classless routing. BGP is defined in RFC 1771, “A Border Gateway Protocol 4 (BGP-4)” and RFC 1772, “Application of the Border Gateway Protocol in the Internet” in the IETF RFC Database.
  
  

The Windows Server 2003 Routing and Remote Access service does not include EGP or BGP.

IGPs are intra–autonomous system routing protocols. Routers use an IGP to forward routing information to the other routers within the autonomous system. IGPs can distribute routes within the autonomous system in either a flat or hierarchical manner.

The IGPs for IP internetworks include:

• RIP for IP. An RFC-based distance vector IGP.
  
  
• OSPF. An RFC-based link state IGP.
  
  
• IGRP. A distance vector IGP developed by Cisco Systems, Inc.
  
  

The Windows Server 2003 Routing and Remote Access service includes RIP and OSPF, but it does not include a version of IGRP. For detailed information about RIP and OSPF, see the next section, “Common IGP Dynamic Routing Protocols,” and see the companion document How Unicast IPv4 Routing Protocols and Services Work.

A third-party software vendor can develop a version of IGRP that works with the Routing and Remote Access service. For information about Microsoft platform software development kits (SDKs) that provide application programming interfaces (APIs) useful for third-party developers who create software that interacts with Windows Server 2003, including the Routing and Remote Access service, see “IPv4 and Routing and Remote Access APIs” in How Unicast IPv4 Routing Protocols and Services Work.

The most common types of dynamic routing protocols used within a single autonomous system are:

• Distance vector routing protocols
  
  
• Link state routing protocols
  
  

A distance vector routing protocol, such as RIP v2, is more appropriate for a relatively small, simple network that is not expected to grow rapidly. For a large, complex internetwork, a link state routing protocol, such as OSPF, is more appropriate. You must use RIP v2 or OSPF to support VLSM and route summarization. Although the outdated RIP v1 is still widely used in private networks, it does not support either VLSM or route summarization and thus is not well suited for enterprise networks.

Note

The Windows Server 2003 Routing and Remote Access service does not support the dynamic routing protocols RIPng (the version of RIP for IPv6) or OSPF for IPv6. However, a computer running Windows Server 2003 can act as a static IPv6 router, and it is possible to route IPv6 traffic over an IPv4 internetwork using an IPv6 transition technology. For information about Windows Server 2003 support for static IP routing in an IPv6 environment, see “How IPv6 Works” in IPv6 Technical Reference.

Understanding how distance vector and link state routing protocols work is essential to choosing the type of dynamic routing that best suits your network needs. The primary differences between distance vector and link state routing protocols include:

• What routing information is exchanged
  
  
• How the information is exchanged
  
  
• How quickly the internetwork can recover from a downed link or a downed router
  
  

Distance Vector Dynamic Routing Protocols

Link State Dynamic Routing Protocols

Routing problems can occur when either a host routing table or a router routing table contains information that does not reflect the correct topology of the internetwork. On an internetwork that uses dynamic routing, the routing internetwork must reconfigure itself to reflect the new topology whenever a link or router fails. The internetwork is in an unstable state during the period of time it takes for convergence to take place — during this period of time, routing loops and black holes can occur.

Routing loops

During the routing process, packets are forwarded by the sending host and then by one or more routers in the optimal direction as determined by information in each node’s local routing table. If the routing table entries on all routers are correct, a unicast packet takes the best path from the source node to the destination node. However, if any routing table entry is incorrect, either through configuration errors or through learned routes that do not accurately reflect the topology of the internetwork, routing loops can occur.

A routing loop is a path through the internetwork that runs in a circle instead of reaching the intended destination. A routing loop occurs when routers forward traffic to each other in a loop that does not include the network segment of the destination.

The following figure illustrates an example of a routing loop that occurs when inaccurate information is stored in local routing tables.

In this example:

• According to the routing table on Router 1, the optimal route to Network 10 is through Router 2.
  
  
• According to the routing table on Router 2, the optimal route to Network 10 is through Router 3.
  
  
• According to the routing table on Router 3, the optimal route to Network 10 is through Router 1.
  
  

The result would be an infinite loop — except that routable protocols use a counter in the network layer header of the packet to prevent the packet from perpetually looping. Each time a router passes the packet from one network segment to another, the router either increases or decreases the counter. If the count reaches its maximum value (when increasing) or reaches 0 (when decreasing), the packet is discarded by the router. For example, when an IP node sends an IP packet, IP sets a maximum link count in the Time to Live (TTL) field in the IP header. Each IP router crossed by the packet decreases the TTL value in the packet by one. When the TTL value is 0, the IP router discards the packet and sends an ICMP Time Exceeded-TTL Exceeded in Transit message back to the sending node from which the packet originated.

One common type of routing loop can be avoided by ensuring that neighboring routers are not configured with default routes that point to each other: that is, on each router, the default route (0.0.0.0, 0.0.0.0) must not have the Default gateway field configured with the IP address of the other router. Because a default route passes all traffic that is not on a directly connected network to the configured router, routers that have default routes pointing to each other can produce routing loops for traffic with an unreachable destination.

By typing ping -i 255 IP_address at a command prompt, you can initiate the detection of a possible routing loop. The -i parameter sets the TTL value in the ICMP Echo message. If the command results in the message TTL Expired In Transit, this indicates a possible routing loop. By typing tracert IP_address at a command prompt, you can then confirm that a routing loop exists if the output displays a set of repeating router IP addresses.

Black holes

The IP protocol is a connectionless, datagram-based protocol, which, by definition, does not guarantee a successful delivery. IP attempts a best effort, unacknowledged delivery to the next hop or to the final destination, which can lead to conditions on the internetwork in which data is lost.

If a downstream router fails and the failure is not detected by an upstream router, the upstream router continues to forward packets to the failed router. Because the failed downstream router does not receive the packets, the packets forwarded by the upstream router are dropped from the internetwork. The upstream router is said to be sending packets to a black hole, which is defined as a case in which packets on an internetwork are lost without any indication of an error.

In the following figure, Router 1 is not informed that Router 2 has failed and, therefore, continues to forward packets to Router 2. Router 2 is now a black hole because the packets are dropped and no error message is returned to Router 1.

A black hole can form when a link or router fails, and the failure is not yet detected. In a dynamic routing environment, routers detect failed links or failed routers when the lifetime of learned routes expires in their routing tables. In a static routing environment, a black hole persists until the functionality of the link or router is restored or until the static routers are reconfigured by the network administrator.

Black holes can also occur when an active router discards packets without indicating why the packets are discarded. For example, a Path Maximum Transmission Unit (PMTU) black hole router discards IP packets that need to be fragmented without returning a message to the sender indicating the error. A PMTU black hole router silently discards IP packets that require fragmentation if the Don't Fragment (DF) flag in the IP header of the received packet is set to 1. PMTU black hole routers can be difficult to detect because packets of smaller sizes are forwarded.

Link state protocols more robust than distance vector protocols