On 3/31/15, 2:29 PM, "George, Wes" 
<[email protected]<mailto:[email protected]>> wrote:

Wes:

Some replies below.

Thanks!

Alvaro.


Added SIDR, responses below inline, which of course messed up your original 
numbering.
I haven't pushed the rev yet, as I'm waiting to make sure that I don't need to 
make additional changes based on the revision to the BGPSec protocol doc.

. . .

  1.  Section 3. s/SHOULD NOT/should not     We can’t use rfc2119 language to 
mandate the behavior of an operator; in this case related to the amount of time 
the transition phase lasts.  I think you have made a good point about the need 
to not stay there forever, but have also said that the transition is not always 
under the control of the operator.

WG] I disagree with the blanket assertion that we can't normatively mandate the 
behavior of an operator in the context of a standard or BCP document, and I 
think that the guidance is correct here – operators SHOULD NOT stay in 
transition indefinitely. Should rather than must because we know that there are 
extenuating circumstances and acknowledging that we tried to keep it simple.

rfc2119 says, about the keywords: "MUST only be used where it is actually 
required for interoperation or to limit behavior which has potential for 
causing harm”.

While I completely agree that the transition shouldn’t take forever, the use 
that you’re proposing doesn’t have anything to do with interoperability, nor 
will taking longer than expected/usual actually causes harm from the 
specification point of view.  Yes, the operator will have to deal with a 
cumbersome configuration/deployment, but the knobs in this case are specified 
"to keep the additional complexity during the transition period to a minimum”.

To all this, how would you even determine if someone was not adhering to the 
recommendation?  Do you measure the time in days, weeks, months, years?   I’m 
sure that the transition timeframe is very dependent on the 
network/deployment/operations and probably many other things I haven’t even 
thought about.

. . .

  1.  3.2.1 s/MUST/must     You’re making reference to something that is not 
specified in another document..

WG] right, I'm observing that the behavior is not normatively specified, hence 
the caps.

I’m not sure if you’re agreeing with me or if you’re further justifying the use.

The section we’re talking about reads:

3.2.1.  Outbound announcements (PE-->CE)

   When PE1 is moved from AS64510 to AS64500, it will be provisioned
   with the appropriate keys for AS64500 to allow it to forward-sign
   routes using AS64500.  However, there is currently no guidance in the
   BGPSec protocol specification on whether or not the forward-signed
   ASN value MUST match the configured "remote-as" to validate properly.
   That is, if CE1's BGP session is configured as "remote-as 64510", the
   presence of "local-as 64510" on PE1 will ensure that there is no ASN
   mismatch on the BGP session itself, but if CE1 receives updates from
   its remote neighbor (PE1) forward-signed from AS64500, there is no
   guidance as to whether the BGPSec validator on CE1 still consider
   those valid by default.  RFC4271 [RFC4271] section 6.3 mentions this
   match between the ASN of the peer and the AS_PATH data, but it is
   listed as an optional validation, rather than a requirement.
   Assuming that this mismatch will be allowed by vendor implementations
   and using it as a means to solve this migration case is likely to be
   Problematic.

Sounds to me that you want BGPSec to say that the values MUST match, is that 
correct?

If so, then I see two ways forward:

  1.  Work to include that “MUST” in the BGPSec draft.  And then reference it 
here.
  2.  Given that the BGPSec draft deals with the base BGP spec and that here 
we’re talking about the new mechanism from idr-as-migration, then change the 
text above to explicitly say what you want to happen (“MUST match”) and then 
this document would be eventually marked as updating the BGPSec base spec.  [We 
might need to consider the order of processing of the documents through the 
IESG so that the update makes sense.]

. . .

  1.  5.3  "While this is not prohibited by BGPSec 
[I-D.ietf-sidr-bgpsec-protocol], routers that receive updates from iBGP 
neighbors MUST NOT reject updates with new (valid) BGPSec attributes..”  Does 
this represent an update to BGPSec?  Are you implying that the iBGP receiver 
should check the validity of the attributes?  I know that at this point it is a 
little weird to update a draft (not an RFC), but the process should take care 
of the appropriate references. [Maybe the update is not needed based on the 
discussion in the WG today.]

WG] what this is saying is that BGPSec is currently silent on this, and making 
an update to clarify what the behavior needs to be. As to whether the updates 
need to be validated in iBGP, I'm not explicitly requiring that, as I think 
it's really more of an implementation detail.

Ok, so it is an update to BGPSec.  This is important because the header should 
say so and I would like to see a section that summarizes the updates.

As far as the iBGP validation, I think the word “valid” is what was throwing me 
off.  You mean valid as in well formed, etc.
_______________________________________________
sidr mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/sidr

Reply via email to