On Wed, Nov 7, 2012 at 6:38 PM, Eric Osterweil <[email protected]> wrote:
>
> On Nov 7, 2012, at 6:10 PM, Christopher Morrow wrote:
>
>> On Wed, Nov 7, 2012 at 5:47 PM, Eric Osterweil <[email protected]> 
>> wrote:
>>>
>>> On Nov 7, 2012, at 4:11 PM, Christopher Morrow wrote:
>>
>> just as broke, bug lodged (I hope) for a fix.
>
> But, but, it works for me... jk ;)
>

it's (apparently, or hopefully) a display only problem... which causes
pita behavior for the responder (me in this case) if i want things to
sort of look legible.

>>> <broken record>
>>> The offer to provide input to the threats and requirements docs is a genuine
>>> attempt to help.  For real! :)  I have the sneaking suspicion that this 
>>> week's...
>>> err, event... cost some people some sleep.  I feel like we _should_ be
>>> addressing it, no?
>>> </broken record>
>>
>> don't disagree, haven't disagreed. (I think)
>
> I hear ya, but I feel like this obviates our need to begin discussions with 
> new
> design proposals.  How can I offer you a design if we don't agree on what it 
> should do, right?

'it should stop leaks' or as someone else said:  "Show how to know
that the leak is a leak and not another backup path coming to light
for other reasons in the system"

>> nor did filtering on the upstream/customer peerings (clearly)... so
>> folk that want to fall back on 'but de filterz!!' .. are also in
>> failville.
>
> Which of the parties was filtering?

neither it would seem.

>> maybe? folk have to see a benefit to any of this in order for it to
>> move into production use.
>
> Right on...  I think that kind of thinking is really what's motivating a lot 
> of
> leak-based-discussion.  Quick wins might come in plugging leaks, so...
> on to the threats doc? :-P
>
>> if as a first step resource certification
>> work gets me better filtering and/or more people filtering because
>> more reliable filtering is possible, good.
>
> Yeah, we're on the same page w/ resource certification being a prerequisite.
>
>> eventually though, people will realize that:
>>  1) not everyone is filtering
>>  2) not everyone will filter
>>  3) the only solution left is something in the bgp update :(
>>
>> if there is another '3', let's talk about that.
>
> Just to follow the same logic, could I not say,
> 1) not everyone is running bgpsec [clearly, it's not here yet]
> 2) not everyone will run bgpsec [therefore there will be isolated islands]
> 3) the only solution left can't be something in the bgp update?

not sure it quite goes that way, but i (and I think everyone else) do
see that not everyone
in the bright/new/shiney/future future will be happily configured and
using bgpsec. Even if everyone does, it's fairly clear that there will
be a range of answers to "the route signatures dont match".

There's a clear entrance/exit protocol for the 'islands'... this isn't
getting us to a 'what is a leak, in a way I can tell from even a
directly connected peer' though.

>> where else does the data from which you make a decision about 'leak or
>> not' come from?
>
> My understanding is that we need to fallback to intent.  With some way to 
> expose
> and verify intent, we get there.  Discussing how to do that seems like the 
> right first
> step, imho.

ok, see brian's docs... i think.

>>> seem to be enough complexities in the policy semantics that the amount
>>> of effort needed to get the minimum level of expressiveness into BGP
>>> could be a non-starter...  Just my 0.02...
>>
>> perhaps you could explain more/better this problem.
>
> Maybe I could just say, today we have a very complex policy specification 
> language, and
> it's not clear how much of it is used, by whom, and to what extent... If we 
> were to say,
> ``throw it all in an update,'' I think we'd see BGP updates bigger than the 
> bgpsec updates.

why?

>  As of now, if we can't agree that leaks are a problem worth addressing 
> and/or we can't

everyone stop... 'if we can't agree that leaks are a problem worth addressing'

everyone (well, almost everyone I've read) has agreed they are a
problem. Even one worth addressing. There's a path that, follow the
path.

now, re-read the above. please.

> agree on what leaks actually are, how could we _not_ use a very verbose 
> language to
> expose relevant policy?
>
>>>> I think techically rpki is just the jumble of certificate data and roas 
>>>> that talk about:
> <snip>
>> this is less concerning... and it's beside the discussion of leak-or-not.
>
> Yeah, but I didn't bring it...  I just wanted to clarify the above comment.  
> I didn't mean
> to distract, but felt the clarification was important.

ok, it didn't seem to clarify... just distract. on we go though!

>> as noted on-list more than one time I'd love to see some scaling and
>> such work done on the certificate repository... that's not today's
>> problem though.
>
> Yeah, working on that for sure...

excellent.

>>> What are we (as a wg) trying to solve? I.e. shouldn't we be talking about 
>>> threats
>>> and requirements?
>>
>> i think we are/were.
>
> Then I would propose that the insistence on only submitting solutions before
> allowing requirements discussions is not appropriate.

i had thought the requirement was 'stop leaks'.
if so, again: great.
also, again: how.

-chris
_______________________________________________
sidr mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/sidr

Reply via email to