Updated: Never-Ending Story of IP Fragmentation
In mid 2000s I wrote a number of articles describing various TCP/IP features. Most of them are a bit outdated, so I decided to clean up, update and repost the most interesting ones on ipSpace.net, starting with Never-Ending Story of IP Fragmentation.
The first part of that article is already online, covering MTU basics and drawbacks of IP fragmentation.
Maybe you will find it interesting:
- Multicasts and PMTUD - not sure if anything has changed since I last touched the topic (the big multicast packet is silently dropped w/o sending back any ICMP message and it was as RFC stated) and we need to fragment the multicast packets
-if fragmentation is required on the tunnel we can either fragment inner (before encapsulation) or outer packets (encapsulated)- consequences are different.
When inner packet is fragmented the endpoint can reassemble the packets; when outer packet is fragmented the other side of the tunnel is reponsible for it as we need to reassemble packet before de-encapsulating it
- when fragmenting multicast packets - can the endpoint host reassemble them? Always?
- implementation matters and all the kludges vendor have for this
- etc