Inter-VLAN Routing Architectures

Within a LAN topology, inter-VLAN routing is used to route packets between different VLANs. Three common inter-VLAN routing architectures are used in modern LAN networks today:

• Router-on-a-stick

• Router-on-a-stick using trunks

• Layer 3 switching

This section examines each of these in detail, outlining any restrictions or issues associated with each.

Router-on-a-Stick

The router-ona-stick architecture is the most basic method of inter-VLAN routing. In this architecture, a router is simply connected to each VLAN and forwards inter-VLAN traffic between the appropriate VLANs. Figure 5-1 shows this architecture.

Figure 5-1. Router-ona-Stick

As you can see in Figure 5-1, the router has a physical Ethernet interface dedicated for each VLAN. If IP hosts on VLAN 100 need to communicate with hosts of VLAN 200, IP packets with the appropriate source and destination IP addresses are sent to the router, which looks up the destination IP address and forwards to the appropriate host on the destination VLAN. The router-on-a-stick architecture is simple to understand because the Layer 2 functions (provided by a switch) and Layer 3 functions (provided by a router) are physically separated.

The major issue with this architecture is performance. Because routers are software-based, they cannot route packets as fast as switches (hardware-based) can switch frames. Even if you are using high-performance routers, the physical interface connecting each VLAN to the router is a bottleneck because it can transmit packets only at 10 Mbps, 100 Mbps, or 1 Gbps depending on the interface type. This restriction means that the router becomes a performance bottleneck when routing between high-speed VLANs.

Another issue with this architecture is the number of routers and physical interfaces required to support multiple VLANs. A dedicated Ethernet interface is required per VLAN. Routers are low-density devices, meaning that there is a high cost per port and multiple routing devices might be required to support multiple VLANs, increasing the complexity of the network.

Finally, all inter-VLAN traffic must travel via the router. In Figure 5-1, even though the PCs in VLAN 100 and VLAN 200 are connected to the same switch, all inter-VLAN traffic between the PCs must be sent through the router, which is inefficient.

Router-on-a-Stick Using Trunks

As discussed in the last section, the router-on-a-stick architecture has physical limitations based upon a dedicated physical interface being required for each VLAN. This limitation can be removed by using trunk interfaces, where multiple VLANs are supported on a single physical interface by using tagging technologies such as 802.1Q or ISL. Rather than using physical interfaces to attach the router to each VLAN, virtual or logical interfaces are used to attach the router to each VLAN. Figure 5-2 shows this architecture.

Figure 5-2. Router-on-a-Stick Using Trunks

In Figure 5-2, virtual interfaces (rather than physical interfaces) are used to connect the router to each VLAN. A single physical trunk interface transports tagged VLAN traffic to the router, with the tag determining to which virtual interface a frame should be forwarded for routing. Apart from the differences between using physical interfaces per VLAN as opposed to virtual interfaces per VLAN, this architecture is essentially identical to the traditional router-on-a-stick architecture and suffers the same performance limitations, because the routing engine is still software-based and the trunk interface is limited to 10 Mbps, 100 Mbps, or 1 Gbps.

Routing Using Layer 3 Switches

The architectures discussed thus far represent the traditional inter-VLAN routing architectures. The major issue with these architectures is performanceif gigabit speed routing is required between VLANs, extremely high performance and costly routers are required. A new form of inter-VLAN routing on the LAN has emerged in recent years called Layer 3 switching. With a Layer 3 switch, the traditionally separated Layer 2 and Layer 3 functions are combined into a single device, eliminating the bottleneck associated with the cable between a router and switch by replacing the cable with a high-speed backplane connection. Layer 3 switches also typically perform routing in specially designed hardware circuitry rather than software, using specialized hardware that can perform routing functions at high speed. This means that the performance of Layer 3 switches is much higher than traditional router-on-a-stick architectures. For example, if you use a Cisco 3640 series router in the router-on-a-stick architecture, you can achieve routing speeds of up to 40,000 packets per second. If you compare this with a Cisco Catalyst 3550-24-EMI Layer 3 switch, which is actually cheaper than a Cisco 3640 router, you can route packets at up to 6.6 million packets per second. This is obviously quite a difference and highlights the limitations of using router-on-a-stick architectures for inter-VLAN routing on the LAN. Of course, the Cisco 3640 router still has a place in the network; it supports a wide variety of diverse media, including serial and ATM connections for WAN connectivity; also supports advanced features such as firewalling, encryption, and so onall of which are not supported on Cisco Catalyst switches.

The Layer 3 switch uses application-specific integrated circuits (ASICs), which are hardware chips that can route traffic at very high speeds. These ASICs are installed on the switching engine of a Layer 3 switch, which traditionally switches frames at Layer 2. The ASICs allow the switching engine to also switch frames that contain packets sent between different VLANs. Each ASIC is programmed with the information required to route traffic from one VLAN to another, without having to pass the traffic through the CPU of the routing engine. This information includes the egress port, egress VLAN, and new destination MAC address that should be written for the frame that is sent. Some form of route cache is normally used to store such information, with the ASIC searching the cache for routing information for the destination IP address of packets as they are received. How this information is programmed into the route cache depends on the Layer 3 switch architecture used; however, the end result is essentially the same.

In addition to the high-speed routing feature, these ASICs also can apply security access control list (ACL) filtering and Layer 3 quality of service (QoS) classification, all at wire-speed, meaning these useful features can be turned on without affecting performance.

NOTE

The internal mechanics of Layer 3 switching are covered in more detail in Chapter 6, "Layer 3 Switching."

When examining the architecture of a Layer 3 switch, it is important to understand that several different approaches to Layer 3 switching implemented by Cisco exist:

• Router-on-a-stick Some chassis-based Catalyst switches (e.g., the Catalyst 4000 and 5000) support routing modules, which are effectively routers on a blade. Apart from having a high-speed connection to the switch backplane, the routing module is essentially a router-on-a-stick, with all routed traffic requiring processing through the routing module. This architecture is not really Layer 3 switching at all because the switch hardware has no special ASICs for Layer 3 switching; instead, it is a high-speed, router-on-a-stick architecture.

• Multilayer switching (MLS) In this architecture, hardware-based ASICs on the switching component of the Layer 3 switch refer to a cache that is populated with the required information to route a packet received on one VLAN to another VLAN, without having to pass the packet through the routing engine. With MLS, the Layer 3 switching cache is populated after the first packet of a particular flow (connection) is received and the route processor is queried for routing information.

• Cisco Express Forwarding (CEF) This architecture is identical to MLS in terms of the hardware-based ASICs referring to a Layer 3 cache for information as to how to route packets between VLANs without involving the router processor. CEF differs from MLS in terms of the way the Layer 3 cache is populated. CEF pre-populates the caches with full routing information, which means the route processor never needs to be queried for the initial routing information that is required in a MLS architecture.

In this chapter, you learn how to configure the Catalyst 4000 using the Layer 3 routing module in a router-on-a-stick architecture. You also learn how to configure Layer 3 switching on the Catalyst 3550, which is based upon a CEF architecture. In Chapter 6, you learn about MLS and CEF on the Catalyst 6000/6500 family of switches.