Minimizing the Impact of Control Plane Traffic on the CPU

In many cases, you can disable the reception and transmission of certain types of packets on an interface to minimize the amount of CPU load that is required to process unneeded packets. These types of packets fall into a category known as process switched traffic. This traffic must be handled by the CPU and hence results in a performance impact on the CPU of the network device.

Process switched traffic falls into two primary categories:

Receive adjacency traffic: This traffic contains an entry in the Cisco Express Forwarding (CEF) table whereby the next router hop is the device itself, which is indicated by the term receive in the show ip cef command-line interface (CLI) output. This indication is the case for any IP address that requires direct handling by the Cisco IOS device CPU, which includes interface IP addresses, multicast address space, and broadcast address space.

Example 13-1 provides sample output generated when issuing the show ip cef command. Any of the IP addresses/subnets for which “receive” is listed as the Next Hop indicates that packets destined for this address space will end up hitting the control plane and CPU.

Example 13-1 show ip cef Output

Click here to view code image

CCNA-Router-1# show ip cef
Prefix               Next Hop             Interface
0.0.0.0/0            no route
0.0.0.0/8            drop
0.0.0.0/32           receive
10.2.2.0/24          192.168.10.2         GigabitEthernet0/1
127.0.0.0/8          drop
192.168.10.0/24      attached             GigabitEthernet0/1
192.168.10.0/32      receive              GigabitEthernet0/1
192.168.10.1/32      receive              GigabitEthernet0/1
192.168.10.2/32      attached             GigabitEthernet0/1
192.168.10.255/32    receive              GigabitEthernet0/1
192.168.15.0/24      attached             Loopback1
192.168.15.0/32      receive              Loopback1
192.168.15.1/32      receive              Loopback1
192.168.15.255/32    receive              Loopback1
192.168.30.0/24      192.168.10.2         GigabitEthernet0/1
192.168.100.0/24     attached             Loopback0
192.168.100.0/32     receive              Loopback0
192.168.100.1/32     receive              Loopback0
192.168.100.255/32   receive              Loopback0
192.168.200.0/24     192.168.10.2         GigabitEthernet0/1
224.0.0.0/4          drop

CCNA-Router-1#

Data plane traffic requiring special processing by the CPU: Although this chapter focuses on control plane traffic, there is still a need to address certain data plane traffic (that is, traffic which has a destination beyond, or through, the network device in question). The following types of data plane traffic require special processing by the CPU resulting in a performance impact on the CPU:

Access control list (ACL) logging: ACL logging traffic consists of any packets that are generated due to a match (permit or deny) of an access control entry (ACE) on which the log keyword is used.

Unicast Reverse Path Forwarding (Unicast RPF): Unicast RPF, used in conjunction with an ACL, can result in the process switching of certain packets.

IP options: Any IP packets with options included must be processed by the CPU.

Fragmentation: Any IP packet that requires fragmentation must be passed to the CPU for processing.

Time-To-Live (TTL) expiry: Packets that have a TTL value less than or equal to 1 require “Internet Control Message Protocol Time Exceeded (ICMP Type 11, Code 0)” messages to be sent, which results in CPU processing.

ICMP unreachables: Packets that result in ICMP unreachable messages due to routing, maximum transmission unit (MTU), or filtering are processed by the CPU.

Traffic requiring an ARP request: Destinations for which an ARP entry does not exist require processing by the CPU.

Non-IP traffic: All non-IP traffic is processed by the CPU.