RE: Meta-issues: On the deprecation of the fragmentation function

[RJ Atkinson <[e-mail]>]

All,

I think everyone agrees that intermediate fragmentation
(e.g. by routers having high-speed links in the middle of the path) 
can be problematic,and that is why IPv6 does NOT require that 
(i.e. RFC-2460 says that the "DF" bit is invisible but always 
set for IPv6).

That noted, I agree very much with Fernando Gont's
comments on 5th July 2013 that IPv6 fragmentation in
end systems ought NOT be deprecated -- and I find his
reasoning compelling.


(SEPARATELY)

Not all deployed links are built out of wired/wireless
Ethernet or SONET.  The assumption that everything is either
Ethernet or SONET (or perhaps WDM) is a bit of a "First
World" assumption and puts lesser-developed countries/regions
at a significant disadvantage for Internet access.  

Frankly it is embarrassing for IETF participants to make 
such an assumption (i.e. that all links are Ethernet, 
SONET, WDM, or better).

While it is fine to optimise so that Ethernet or SONET links 
have great performance, for example by eliminating the
requirement for backbone routers to fragment packets,
we ought to retain the ancient  "IP OVER EVERYTHING"  mantra 
that has served the Internet (and ALL regions of the world) 
so well for so many years now.

I know of a number of link technologies that simply can't 
support the existing 1280 byte minimum IPv6 MTU.  Links 
with the lower MTU overwhelmingly are relatively low speed 
wireless links (everything in the gamut from HF radio 
to SATCOM).  My concerns about this have been expressed in
IPv6 WG meetings dating back to the early 1990s, as IETF
meeting minutes show.  All the data available to me indicates 
that these lower-speed (and smaller MTU) links are NOT going away.

For HF radio links, link transmission speed is usually 
measured in Kbps, with some deployed IPv4/HF links today
running as low as 2.4 Kbps.  VHF and UHF line-of-sight
links tend to be much faster than HF, but are still well 
below the slowest modern Ethernet link (i.e. 10 Mbps).

The IPv6 concept of adding a special link-specific
fragmentation/reassembly protocol layer has never been
practical.  By definition, these links have severe bandwidth
limitations already.  I've seen multiple instances where
specialised routers that are connected to these LOW speed
links insert IPv6 Fragmentation Headers and perform
intermediate fragmentation to get IPv6 packets through
such links.  In these special cases, there is no performance
issue because these wireless links have such low data rates.

A result of deprecating the IPv6 Fragmentation Header will be
that such links will be running IPv4 *forever*.  This is
the opposite of encouraging IPv6 transition -- it is preventing
IPv6 transition.  It also probably requires the ongoing deployment 
of IPv6::IPv4 protocol translation gateways -- and also
probably inhibits Internet deployment outside of "1st World"
countries.

Sigh.


Deprecating IPv6 Fragmentation Header and related end-system
support is going to harm users and inhibit transition to IPv6.


PLEASE, PLEASE, let us NOT go there.


Sincerely Yours,

Ran Atkinson