Path MTU Discovery

Sending the largest possible packets maximizes efficient use of network capacity when bulk data transfers are performed. Because IPv6 routers no longer support fragmentation, the sending host must either fragment its payload (not recommended) or discover the maximum-sized packet that can be sent to the destination and send unfragmented packets at that size.

The path maximum transmission unit (PMTU) is the smallest link MTU supported by any link in the path between a source and a destination. The link MTU is the maximum-sized link-layer payload that can be sent on the link. This corresponds to the maximum-sized packet that can be sent on the link, but differs from the maximum-sized frame that can be sent on the link. The maximum-sized frame includes the link-layer header and trailer. For example, for Ethernet links using Ethernet II encapsulation, the link MTU is 1,500 bytes and the maximum-sized frame is 1,526 bytes (which includes the Ethernet preamble, source and destination addresses, the EtherType field, and the Frame Check Sequence field). For more information about the Ethernet II header and trailer, see Appendix A.

IPv6 packets with a maximum size of the PMTU of the current path do not require fragmentation by the sending host and are successfully forwarded by all routers on the path. To discover the PMTU of the current path, the sending node relies on the receipt of ICMPv6 Packet Too Big messages. The PMTU is discovered through the following process:

1. The sending node assumes that the destination PMTU is the link MTU of the interface on which the traffic is being forwarded.
2. The sending node sends IPv6 packets at the assumed PMTU size.
3. If a router on the path is unable to forward the packet because it is using an interface with a link MTU that is smaller than the size of the packet, it sends an ICMPv6 Packet Too Big message back to the sending node and discards the IPv6 packet. The ICMPv6 Packet Too Big message contains the link MTU of the interface on which forwarding failed.
4. The sending node sets the new assumed PMTU for packets being sent to the destination to the value of the MTU field in the ICMPv6 Packet Too Big message.

The sending node starts again at step 2 and repeats steps 2 through 4 for as many times as are necessary to discover the PMTU. The PMTU is determined when either no additional ICMPv6 Packet Too Big messages are received or an acknowledgment or response packet is received from the destination.

In RFC 1981, it is recommended that IPv6 nodes support PMTU discovery. Those that do not must use the minimum link MTU of 1,280 bytes as the PMTU for all destinations.

Changes in PMTU

Due to changes in routing topology, the path between source and destination might change over time. When a new path requires a lower PMTU, the PMTU process described in the "Path MTU Discovery" section begins at step 3 and repeats steps 2 through 4 until the new PMTU is discovered.

Decreases in path MTU are immediately discovered through the receipt of ICMPv6 Packet Too Big messages. Increases in path MTU must be detected by the sending node. As described in RFC 1981, the sending node can attempt to send a larger IPv6 packet after a minimum of 5 minutes (10 minutes are recommended) upon receiving an ICMPv6 Packet Too Big message.

Figure 5-8 summarizes the PMTU discovery process of a node using the IPv6 protocol for the Windows .NET Server 2003 family.

Figure 5-8. The PMTU discovery process