As already stated in 3. IPv6 primary differences from IPv4, IPv6 does not allow intermediate routers to fragment packets. Instead, IPv6 pushes the responsibility of fragmentation to the source node. If a packet exceeds the MTU, it is either fragmented by the sender or dropped. This means the sender must use PMTUD [RFC7690] to ensure that packets are sized appropriately for the smallest MTU along the path. The sender fragments packets if necessary before sending them. This require additional computation on the sender to fragment packets but there has been no significant performance implication reported.
A jumbo frame is an Ethernet frame that is larger than the standard 1500 bytes, commonly configured to be around 9000 bytes. If a router along the path has a smaller MTU when sending jumbo frames in an IPv4 network, it will fragment the frame. This can lead to higher fragmentation overhead because the larger the original frame, the more fragments it must be split into. Additionally, fragmentation adds processing complexity at both the router and the destination where reassembly occurs.
IPv6 avoids this fragmentation overhead by relying on PMTUD [RFC7690]. If a jumbo frame exceeds the MTU of any network hop, the sender is responsible for fragmenting it before transmission. However, if properly configured, the sender can send larger packets efficiently without fragmentation, provided that the entire path supports jumbo frames. Ths allows IPv6 to handle larger packets more effectively because the Path MTU Discovery mechanism ensures that packets fit within the MTU of every hop along the route. This mechanism is defined in [RFC8201].
The “Jumbo Payload Option” in IPv6 [RFC2675] allows packets larger than 65,535 bytes (the maximum payload size for standard IPv6 packets) to be transmitted. This option is included in the Hop-by-Hop Options header and enables IPv6 to support super jumbo frames efficiently, even when dealing with extremely large packet sizes. This mechanism simplifies the handling of large packets without requiring them to be split into smaller fragments. If a network supports large enough MTUs, IPv6 can use this option to transmit large frames without intermediate fragmentation. However, it is very little used because it needs a layer 2 technology supporting very big packets. An interesting use case is for internal communication in support of segmentation offload, described in this blog entry.