The enhanced packet core (EPC) defines many different interfaces that require encapsulation of IPv4 and IPv6 data packets. Because the EPC adds encapsulating headers, additional care must be taken when fragmenting IPv4 and IPv6 packets.

Appropriate configuration should not result in fragmentation at any node in EPC. This feature fragments the IPv6 and IPv4 packets based on their MTU.

In RFC-4861 there is a provision to send the Maximum Transmission Unit (MTU) in Router Advertisement (RA) messages. P-GW supports the sending of the IPv6 MTU option in RAs for IPv6 and IPv4v6 PDN types towards the UE. The (Internet) can now send downlink data packet and based on the configured MTU, data fragmentation is performed at the source, if required. This feature also reduces the number of ICMPv6 Packet Too Big Error messages in the customer's network.

The MTU size is configurable through the Command Line Interface (CLI) on P-GW.

The following CLI commands configures the Maximum Transmission Unit (MTU) for data sent on the IPv4 and IPv6 tunnel between the P-GW and the mobile node:

configure 
   context context_name 
      apn apn_name 
         ppp mtu bytes 
         data-tunnel mtu bytes 
         policy ipv6 tunnel mtu exceed { fragment inner | notify-sender | fragment } 
         access-link ip-fragmentation { df-ignore  | normal | df-fragment-and-icmp-notify } 
         end 

NOTES:

• bytes: Specifies the MTU for the IPv6 tunnel between the P-GW and the mobile node. bytes must be an integer between 1280 and 2000. Default: 1500.

• ppp: Specifies data sent on the IPv4 tunnel between P-GW and mobile node.

• data-tunel mtu: Specifies data sent on the IPv6 tunnel between P-GW and mobile node.

• fragment inner: Performs one time fragment at GTP tunnel initiator.

• notify-sender: System will drop the incoming packet and send "ICMPv6 Packet Too Big" to the original sender.

  Note	This is also the default CLI configuration, hence this should be the default behavior when nothing is explicitly configured.

• fragment: Performs fragmentation or reassembly at intermediate GTP hops.

• df-ignore: Ignores the DF (Don't Fragment) bit setting; fragments and forwards the packet over the access link.

  Note	This is also the default CLI configuration, hence this should be the default behavior when nothing is explicitly configured.

• df-fragment-and-icmp-notify: Partially ignores the DF bit; fragments and forwards the packet, but also returns an ICMP error message to the source of the packet. The number of ICMP errors sent like this is rate-limited to one ICMP error packet per second per session.

• normal: Drops the packet and sends an ICMP unreachable message to the source of packet.