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

[Kristian Larsson]

On Tue, Jul 22, 2008 at 04:21:08PM +0100, Bruce M Simpson wrote:
> We're talking about a code delta which is relatively large compared to 
> asking people to adapt a little to get their IP routing thing for 
> potentially less money.

A code delta which is very easy to produce with
sed and friends. Though you are kind of missing my
point...

> At the moment, "it ain't broke", so let's just leave it for now, IMO.

I work in the Internet service provider business
and have done so for the last couple of years. I
spend roughly half my day staring at an IOS prompt
and I know a great deal of people just like me.
What I don't know, is someone using XORP. A lot of
people have heard about it and many have tried it
out but rejected it for some reason.
Now, I don't only stare at this IOS prompt for a
living, I'm also quite interested in it on a
personal level and spend a considerable amount of 
my free time tinkering with network stuff. Although
my budget is nowhere near that of my employer I have
managed to gather a quite nice pile of equipment
ranging from GSRs to M5s. Over time I have learn
that GSRs consume a shitload of electricity and
puts out an almost equal amount of heat not to
mention the space they occupy - and finding fast
interface for a Juniper is a bi***. This is where
I imagine XORP and other routing suites would step
in. A lot of people pursue the
open-source-routing-suite-path for slightly
different reasons than I but I believe the reasons
for not choosing XORP are the same. More on that 
in a bit...

What I would like is of course to find an open and
extensible router platform to suit my needs. When
I was unable to do so, I instead looked at what
would be the closest match and try improve on that.
Et voila... XORP!
The ambition of course being to help XORP into 
world domination. Not being a programmer has
proven to be somewhat of an obstacle - I've spent
countless hours going through XORP code and
eventually just giving up on whatever I had set
out to do. Coding isn't my ballgame, so what can I
do to help?
Well, I know what I like about routers and I know
what I don't like, perhaps sharing this could help
the project gain some acceptance amongst other
networkers like me.

Hasso Tepper wrote an excellent piece 
http://hasso.linux.ee/doku.php/english:network:xorpsucks
I'm just going to give a +1 on the whole thing.

The lack of support for recursing routes and with
it the absolutely silly side-effect of having to
specify next-hop for your BGP neighbors make the
BGP module close to worthless for iBGP relations.
No templating or even per-peer policies is
hilarious, how do you expect anyone to be able to 
use this in a real network?
I have routers with hundreds of peers, keeping one
policy for all those up to date would be close to
impossible. At least so difficult I don't even
want to think about trying.
As Hasso mentions the CLI displays things in an
awful manner. No leaf-nodes without value or the
mandatory {} for all branch-nodes makes it painful
just to look at the configuration.
The only thing that I can come up with from the
top of my head that I would list as a nice feature
is the marking of new lines (">") in the config.
.. well, then you have the multicast protocols as
well, but XORP doesn't have much competition
amongst the open source alternatives in that area.
Competing against commercial vendors we are down
to the ">"-marking part.

I got carried away a bit, returning to the [46]
thingy...

If we go back to when this was implemented, why
was ospf4 and ospf6 chosen in the first place? Why
wasn't the already familiar ospf[v3] nomenclature
chosen? Why have a Junos-similar shell but deviate
for no apparent reason? 
I've hardly seen a discussion on a list _before_
the name for something was chosen, rather the
implementor chooses something and then it sticks.
For this reason I would like to know the plan for
future "stuff" and maybe even have a saying in it.

I would be sad to see another deviation from what
is the de facto syntax out there. It makes
networking kind of guys dislike the software and
that is probably the worst thing for an open
source routing suite.

It has been argued before that the user base of
XORP is familiar with X and changing X would
create a disturbance and thus it is left as is.
Does XORP have an active user base (more than the
handful of people on this list) ? For how many
people would changing something become a nuisance?

Why not try to adhere to what people are used to
and try to get the basic stuff like CLI and BGP
to work before implementing fancy stuff like
firewall rules? Not that the latter is a bad
thing, I'd just like to see a different
prioritization here. 

Kind regards,
  Kristian.

PS. I'm still impressed by XORPs architecture
(creds for that) though that doesn't change the 
fact that I as a network engineer hate to use
it. DS.

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