Re: Non-Last Small IPv6 Fragments

[Nick Hilliard <[e-mail]>]

Simon Hobson wrote on 14/01/2019 16:00:
> Nick Hilliard <[e-mail]> wrote:
>> No intermediate node is going to fragment anything other than the
>> last frame to 8 bytes, or at least if it does, the network is so
>> severely crippled that fragmentation is the least of its problems.
>> It would be fair to assume that if host B receives a stream of
>> fragments where the fragment size is unfeasibly small, then the
>> fragmentation was done by host A.
> 
> What if ... An intermediate node gets a packet inbound of N bytes,
> but the MTU downstream only allows it to send N-8. If it doesn't have
> other packets* to degragment/merge with first, then it's only option
> is to send two smaller packets downstream - and one logical method
> would be to send N-8 followed by 8 bytes.

I was thinking about a variation of this situation earlier today, e.g. 
1492 MTU DSL products with jumbo transmitter, or possibly frames 
transmitted / load balanced over a different physical path to the path 
that PMTU first figured out.

In summary: tiny fragments which aren't last in the sequence shouldn't 
be expected often, but they form a legitimate corner case.

Nick