Well, does the source node sending an IP packet worry about fragmenting the data being sent, just because some nth hop’s data link does not support such large packets? The source node does not even know about the path the packet is going to take – that’s the essence of connectionless network layer anyways. What if the path the packet changes subsequently?
The original notion of fragmentation with IPv4 made sense to me. If an arbitrary router find its data link to parochial, it will fragment it and send it across. The destination node will re-assemble the packet and deliver it to the gods.
However, IPv6 made source based fragmentation mandatory. The only fact that comes to the source node’s rescue – in dealing the MTU of the path – is the minimum MTU that needs to be supported which is 1280 bytes. If the source node does not wish to participate in this exercise of MTU discovery, it can choose to send the minimum MTU packets safely. Otherwise, it has to run along hops to discover the MTU.
The only reason I can think of for removing this is to relieve routers of online (i.e. data path speed) fragmentation of packets which may need complex hardware to achieve the same.
I did not like this approach.