[Xorp-hackers] OSPF[46]...

[Kristian Larsson]

Dear hackers,

I would like to raise the question on names and
syntax regarding IPv4 vs IPv6 in xorpsh.
As it stands today, many of the commands use a
postfix of either 4 or 6 to denote which
address-family they belong to. In the case of
OSPFv2 and OSPFv3 it maps in nicely, since OSPFv2
only does IPv4 and OSPFv3 only does IPv6. OSPFv3
have the ability, thanks to a TLV based format, to
support IPv4 as well. If that would be the case,
what would be your way to proceed?
Looking at IS-IS, for which development will 
hopefully begin soon, what syntax will be used for
IS-IS as it can handle both IPv4 and IPv6?
Will isis4 be used for commands regarding v4 and
isis6 for v6?

Looking at IOS there's an address-family command
for IS-IS and for ospf it's called 'router ospf'
and 'ipv6 router ospf'. Now, IOS was invented in a
time when IPv6 and thus OSPFv3 wasn't around. If
we instead take a look at the two dominating
router OSes that were designed recently enough for
v6 to exist (XR and JUNOS) they both aptly name it
'ospf' and 'ospfv3'.

The [46] notation fits in perfectly for the fea or
plumbing parts but when used for ospf or some
other protocol it tries to imply limits that's not
really there. Abstracting away routing protocols 
is not something network administrators 
appreciates.
Principle of least astonishment is pretty clear on
this I believe. People used to XORP expect
ospf[46] while everyone else coming from some
platform not being XORP would expect (ospf[v3]).
As XORPs user base is rather limited (btw, does
anyone have any figures?), I think it's clear that
ospf[v3] is the way to go, especially as it eases
transition from other platforms.

An easy transition path would be to add an alias
for ospf4 to ospf and likewise for ospf6 to
ospfv3, possibly adding a warning that ospf[46] is
deprecated somewhere in the CLI.

What's your take on this?

Kind regards,
  Kristian.

PS. I realized I dug real deep into the ospf[46]
thing here while I really wanted to bring up the
general question of separating routing protocols
this way into address-families. It would apply to
ISIS, OLSR or some other protocol as well... DS.

-- 
Kristian Larsson                                        KLL-RIPE
Network Engineer / Internet Core        Tele2 / SWIPnet [AS1257]
+46 704 910401			              kll at spritelink.net