Intro to Cisco Multi Layer Switching and Cisco Express Forwarding

This article is intended for those new to Cisco Express Forwarding (CEF) and its impact on the way Multilayer Switching (MLS) is done in Cisco hardware. Of course, this article can also serve as a review for those familiar with the concepts but are looking for a refresher. In this first article we are going to go over the components that make up this switching architecture followed by some fundamental examples to illustrate these components and concepts at work. Before we get started be sure to download the topology we are going to be using in the lab examples for clarity.

Modern Catalyst Multilayer switches utilize CEF-based MLS. The terminology and architecture of this switching model can be tough to understand at first, but trust me, it really isn't that difficult to grasp after you start working with it.

There are two distinguished functions provided by a CEF-based Multilayer Switch. The first function is building routing information. This routing information is built by the Layer 3 engine within the control plane. The second function provided is hardware switching of packets. Hardware switching of packets is done by the Layer 3 Forwarding Engine within the data plane. The data plane is where CEF works its magic. The control plane is where layer 3 decisions are made, when those layer 3 packets can NOT be switched in hardware.

Since CEFs magic is provided in the data plane, we will start with it. It is the most fun anyway. The Layer 3 Forwarding Engine within the data plane has two components of its own.

The first component is the CEF Forwarding Information Base, and the second is the CEF Adjacency table. The CEF Forwarding Information Base (FIB) is basically just a reformatted routing table ordered such that the most specific routes are found first. The FIB contains next hop information for each prefix. The routing and next-hop information is built in software in the control plane, and then passed to the Layer 3 forwarding engine and placed in the FIB. I can't stress enough how important it is to understand that this is basically a reordered routing table with some additional entries in it. When a packet enters the switch, the switch consults the FIB and finds the longest match prefix and obtains the next hop address. I know this doesn't sound like magic yet, but stay with me, there is more and this stuff is slick. This stuff is why Cisco rules.

The second component, the adjacency table, contains and maintains layer 2 addresses for every entry in the FIB. This table is built the same way the FIB is built. It is built from the ARP table that is built with the Layer 3 engine in the control plane and then passed to the Layer 3 Forwarding Engine and placed in the CEF Adjacency table. If you know how packets are encapsulated and rewritten as they make their way across a layer 3 network, you are probably beginning to develop an idea of what is going to happen with the adjacency table.

Since the FIB and Adjacency tables are both handled in hardware, we're beginning to see how CEF can dramatically improve the performance of layer 3 forwarding operations. It copies the work the Layer 3 Engine does in software, and the Layer 3 Forwarding Engine uses it to make multilayer switching decisions in hardware. Between the FIB having next hop layer 3 information, and the adjacency table having both the layer 3 and layer 2 information, CEF has at its disposal everything it needs to forward packets without consulting a routing table running in software, and without the need to do an ARP for layer 2 header rewrite. It is all in hardware and it all happens at line speed. Don't you love it when tidbits of information all come together? I sure do.

Now, let's take a look at two scenarios to see the paths packets take through a CEF-enabled multilayer switch. In scenario 1, we have a valid FIB entry and associated adjacency table entry. A packet comes in the ingress interface, the FIB is consulted and an entry is found. The FIB is matched on the longest prefix. The layer 2 information is retrieved from the adjacency table and the packet is then forwarded through the packet rewrite engine, which rewrites the appropriate packet and frame header information at line speed and sends the packet out the egress interface. Notice that no ARP requests are made, no software based processing is performed, and frame information is written in hardware.

In scenario 2, a packet comes ingress on an interface, the FIB is consulted and is not able to be CEF switched because of one of several different reasons. At this point the packet is punted to the Layer 3 engine for further processing. We aren't going to cover all the scenarios in which a CEF Punt occurs here. We'll save those for Part II.

It should be apparent, but it is worth mentioning here for clarity. When changes happen in the routing and ARP tables that are maintained by the Layer 3 Engine, those changes are automatically propogated to the Layer 3 Forwarding Engine. This updates the FIB and the Adjacency tables instantaneously.

Now that ALL of that is out of the way, let's start looking at the relationship between the routing table, arp table, the cef fib table, and the cef adjacency table. Let's start by having a look at the IP addresses of the connected interfaces of the two devices used in these demonstrations.

MPLS1#show ip interface brief

Interface                  IP-Address      OK? Method Status                Protocol

FastEthernet0/0            unassigned      YES NVRAM  administratively down down    

FastEthernet1/0            172.16.13.1     YES NVRAM  up                    up      

FastEthernet1/1            172.16.12.1     YES NVRAM  up                    up      

FastEthernet2/0            172.16.15.1     YES NVRAM  up                    up      

FastEthernet2/1            unassigned      YES NVRAM  administratively down down    

FastEthernet3/0            unassigned      YES NVRAM  administratively down down    

FastEthernet3/1            unassigned      YES NVRAM  administratively down down    

Loopback0                  10.0.0.1        YES NVRAM  up                    up      

Tunnel7                    10.0.0.1        YES TFTP   up                    down    

Tunnel702                  10.0.0.1        YES TFTP   up                    down    

Tunnel703                  10.0.0.1        YES TFTP   up                    down    

MPLS2#show ip interface brief

Interface                  IP-Address      OK? Method Status                Protocol

FastEthernet0/0            unassigned      YES NVRAM  administratively down down    

FastEthernet1/0            172.16.12.2     YES NVRAM  up                    up      

FastEthernet1/1            172.16.23.2     YES NVRAM  up                    up      

FastEthernet2/0            172.16.24.2     YES NVRAM  up                    up      

FastEthernet2/1            172.16.25.2     YES NVRAM  up                    up      

FastEthernet3/0            unassigned      YES NVRAM  administratively down down    

FastEthernet3/1            unassigned      YES NVRAM  administratively down down    

Loopback0                  10.0.0.2        YES NVRAM  up                    up      

Now we are going to look at the routing table on MPLS1:

MPLS1#show ip route

Codes: C - connected, S - static, R - RIP, M - mobile, B - BGP

       D - EIGRP, EX - EIGRP external, O - OSPF, IA - OSPF inter area 

       N1 - OSPF NSSA external type 1, N2 - OSPF NSSA external type 2

       E1 - OSPF external type 1, E2 - OSPF external type 2

       i - IS-IS, su - IS-IS summary, L1 - IS-IS level-1, L2 - IS-IS level-2

       ia - IS-IS inter area, * - candidate default, U - per-user static route

       o - ODR, P - periodic downloaded static route



Gateway of last resort is not set



     172.16.0.0/28 is subnetted, 6 subnets

O       172.16.24.0 [110/2] via 172.16.12.2, 01:12:32, FastEthernet1/1

O       172.16.25.0 [110/2] via 172.16.12.2, 01:12:32, FastEthernet1/1

O       172.16.23.0 [110/2] via 172.16.12.2, 01:12:32, FastEthernet1/1

C       172.16.12.0 is directly connected, FastEthernet1/1

C       172.16.13.0 is directly connected, FastEthernet1/0

C       172.16.15.0 is directly connected, FastEthernet2/0

     10.0.0.0/32 is subnetted, 2 subnets

O       10.0.0.2 [110/2] via 172.16.12.2, 01:12:32, FastEthernet1/1

C       10.0.0.1 is directly connected, Loopback0

...And now the FIB on MPLS1. Take note of the similarities and in particular the next hop addresses.

MPLS1#show ip cef

Prefix              Next Hop             Interface

0.0.0.0/0           drop                 Null0 (default route handler entry)

0.0.0.0/8           drop

0.0.0.0/32          receive

10.0.0.1/32         receive

10.0.0.2/32         172.16.12.2          FastEthernet1/1

127.0.0.0/8         drop

172.16.12.0/28      attached             FastEthernet1/1

172.16.12.0/32      receive

172.16.12.1/32      receive

172.16.12.2/32      172.16.12.2          FastEthernet1/1

172.16.12.15/32     receive

172.16.13.0/28      attached             FastEthernet1/0

172.16.13.0/32      receive

172.16.13.1/32      receive

172.16.13.15/32     receive

172.16.15.0/28      attached             FastEthernet2/0

172.16.15.0/32      receive

172.16.15.1/32      receive

172.16.15.15/32     receive

172.16.23.0/28      172.16.12.2          FastEthernet1/1

172.16.24.0/28      172.16.12.2          FastEthernet1/1

172.16.25.0/28      172.16.12.2          FastEthernet1/1

224.0.0.0/4         drop

224.0.0.0/24        receive

240.0.0.0/4         drop

255.255.255.255/32  receive

Now we are going to look at the ARP table on MPLS1..followed by the CEF Adjacency table.

MPLS1#show ip arp

Protocol  Address          Age (min)  Hardware Addr   Type   Interface

Internet  172.16.13.1             -   ca00.0bd0.001c  ARPA   FastEthernet1/0

Internet  172.16.12.1             -   ca00.0bd0.001d  ARPA   FastEthernet1/1

Internet  172.16.12.2            73   ca01.0bd0.001c  ARPA   FastEthernet1/1

Internet  172.16.15.1             -   ca00.0bd0.0038  ARPA   FastEthernet2/0

MPLS1#show adjacency detail

Protocol Interface                 Address

TAG      FastEthernet1/1           172.16.12.2(7)

                                   0 packets, 0 bytes

                                   CA010BD0001C

                                   CA000BD0001D8847

                                   TFIB       02:48:53  

                                   Epoch: 0

IP       FastEthernet1/1           172.16.12.2(17)

                                   0 packets, 0 bytes

                                   CA010BD0001C

                                   CA000BD0001D0800

                                   ARP        02:48:53  

                                   Epoch: 0

The correlations here should all be apparent. Notice the last 4 digits on the line under the bolded MAC addresses. These are ethertype codes. 8847 is MPLS-IP. 0800 is Ethernet.


That about brings Cisco Express Forwarding Part I to a conclusion. That should provide you with a foundational knowledge of what CEF does and how it works. There are quite a few more details to be covered in later articles. Right now I just to get this introduction out there because we will be needing it for MPLS Part III.