Stewart,
I didn't intend to relitigate IPv6 design decisions, but only to refute the misinformation that the choice of address size and format was not the result of technical reasoning. But since you rang the doorbell, see below.
On 02-Nov-21 09:06, Stewart Bryant wrote:
Sent from my iPad
On 1 Nov 2021, at 19:49, Brian E Carpenter <brian.e.carpenter@xxxxxxxxx> wrote:
Somebody whose email never reaches my inbox alledgedly said:
> IPv6 with unnecessarily lengthy 16B addresses without valid
> technical reasoning only to make network operations prohibitively
> painful is a garbage protocol.
Apart from its incivility, this sentence is factually untrue. The address
length was 8 bytes in the early design of what became IPv6, which was of
course essential to overcome the main limitation of IPv4.
There were other techniques of course such as IPv4 in IPv4 which would allowed existing network kit to largely be used unchanged, and would probably have given fewer migration issues.
There were many such proposals around. There's a whole book about it. My personal proposal that would have squared the size of the address space was kicked out after a BOF at the Seattle IETF. There was a pretty extensive process of sieving proposals against requirements.
However you cannot refute a criticism of IPv6 having 128bit addresses by arguing that it started with 64bits because it does actually have 128 bit addresses.
That's not what I did. I explained why we did it.
It was expanded
to 16 bytes when the value of an interface identifier in addition to
a routeable prefix was considered.That idea was based on existing
practice in several non-IP network technologies, and on the IPng
requirements process. In other words, on technical reasoning and on
running code.
Of course variable length and multi-address family approaches existed at the time, and were rejected. One of the reasons for this was politics, in particular a desire not to use an ISO protocol. There was also a counter argument based on forwarder performance, but there were ISO routers that were of comparable speed.
CLNP conforming with US GOSIP had *fixed length* addresses (20 bytes), so
the concern about processing time was not really an argument against CLNP. Incidentally, I came into the IPng process with a strong bias in favour of CLNP, and I had already committed resources to deploying CLNP (DECNET Phase V). But the analysis of solutions against requirements persuaded
me otherwise.
Professionalism includes factual accuracy.
It also includes reviewing if the right decisions were taken so we learn from history.
Agreed.
Brian
