Re: [L2tpext] ppp draft issues - offset padding field

[Carlos Pignataro <[e-mail]>]

Thanks again, please see inline.

On 7/3/2007 3:42 PM, Y Prasad said the following:
> Hi Carlos,
> 
>> -----Original Message-----
>> From: Carlos Pignataro [mailto:[e-mail]]
>> Sent: Thursday, June 28, 2007 7:22 PM
>> To: Y Prasad
>> Cc: Ignacio Goyret; [e-mail]
>> Subject: Re: [L2tpext] ppp draft issues - offset padding field
>>
>>
>>
>> On 6/27/2007 7:31 PM, Y Prasad said the following:
>>> Thanks Carlos for explaining difference between offset size and
> offset
>>> capability. Please see a comment below.
>>>
>>>
>>> -----Original Message-----
>>> From: Carlos Pignataro [mailto:[e-mail]]
>>> Sent: Thursday, June 28, 2007 2:32 AM
>>> To: Y Prasad
>>> Cc: Ignacio Goyret; [e-mail]
>>> Subject: Re: [L2tpext] ppp draft issues - offset padding field
>>>
>>>
>>> Thank you for the comments, please see inline.
>>>
>>> On 6/27/2007 4:25 AM, Y Prasad said the following:
>>>> Hi Carlos,
>>>>
>>>> Please see at <yp>
>>>>
>>>> Rather than a session-level negotiation, how about a tunnel-level
>>>> SCCRQ/SCCRP capability advertisement, much like the "Pseudowire
>>>> Capabilities List" or the "Modem On-Hold Capable AVP" from Section
> 4.1
>>>> of RFC3573. A "Offset Capability" AVP can indicate how big of an
>>> offset
>>>> the sender is willing to insert, if requested at session-level.
>>>>
>>>> <yp> In v3 case tunnel can potentially carry non-ppp sessions. Do
> you
>>>> think offset size useful/needed across all non-ppp sessions as
> well?
>>> This is a good point. Even though v3 supports other PW Types, the
> Offset
>>> Size is only applicable to PPP sessions. Other companion documents
>>> (i.e., other PW Types) do not specify it, and the PPP document
> specifies
>>> other AVPs that are also only applicable to PPP.
>>>
>>>> Common offset size for all the different types of sessions may not
>>>> benefit all the sessions.
>>> Note that there's not necessarily a common offset size for all PPP
>>> sessions. The Offset Size is a Session Management AVP (per-session).
>>> More below.
>>>
>>>> May be we are better off to keep it as session negotiated AVP?
>>> The "OFfset Size" is a Session-requested AVP. The "Offset
> Capability"
>>> only conditions/caps the maximum Offset Size that can be requested
> on
>>> sessions belonging to a given tunnel.
>>>
>>> <yp> wondering what will be the behavior in the following case:
>>> session may get initiated/configured long after the tunnel is up. So
> if
>>> a ppp session that requires offset and the tunnel had not negotiated
>>> offset capability, then we will bring down tunnel to negotiate
> offset
>>> capability?
>> If the offset capability is supported, it needs to advertise it at
>> tunnel creation (SCCRQ, SCCRP). This is true for all capabilities;
> <yp>
> Not sure whether we still need to advertise particular session type
> capabilities (here ppp specific) over the tunnel when the same is again
> negotiated over session. We may have some thing like below

Yes, it would be one or the other. The purpose of the capability
advertisement in SCCRQ/SCCRP is, as you describe below, to not have to
negotiate at the session level, and maintain the same signaling flow as
as with other messages, in which the receiving side requests some
dataplane characteristic.

More importantly, see below re. outgoing calls.

> 
>  Lac  ---------- lns
> ICRQ   ----------> (lac advertise offset capability)
> ICRP   <----------- (lns request offset size and advertise its
> capability)
> ICCN   -----------> lac request offset size
> 
> I am ok for advertising thru SCCRQ/SCCRP though as it helps each session
> not to try for offset negotiation when peer does not support.

OK.
And agreed as another reason.

> 
>>> (IMHO)why not we just mention offset size avp as "ackable" session
> AVP
>>> and avoid offset capability need. I think its ok to define a new avp
>>> type behavior.
>> This is in a way what the current text describes, with the "nak"
>> (meaning "I won't add that offset size") being a CDN. Ignacio argued
>> that control over the offset would need be placed in the device that
>> would add the offset, and that's what the capability solves.
> 
> <yp>  
> Since this addresses only performance issue, un-supported situations may
> not need to end up in disconnecting the session when peer negotiates.
> 
> IMHO, offset padding avp can be a simple ackable non-mandatory avp that
> can be requested by either of the peers independently.
> 
>> There is an additional problem with session-management negotiation of
>> this: In the outgoing call establishment negotiation, the first
> message
>> where a LAC can say "you can request up to this" is with the OCRP, but
>> the LNS has only a chance to request a specific offset size in the
> OCRQ,
>> which happens before. I think that the tunnel-level advertisement
> solves
>> the problem presented by Ignacio in all cases, and is simpler.
> 
> May be this way?
> 
>    LAC  ---- LNS
> 
>        <----- OCRQ (lns advertise offset capability)
>        
>   OCRP  ------> (lac request offset padding and advertise its capability
> as well)
> 
>   OCCN  <-------- (lns reqest offset padding)

The problem here is that the OCCN is in the other direction:
    OCCN  ------>

So if the LAC advertises its capability in the OCRP, the LNS cannot
request the offset padding.

http://tools.ietf.org/html/rfc3931#section-3.4.2
http://tools.ietf.org/html/rfc2661#section-5.2.2

Based on this discussion and your comment above, I'm inclined to update
with an SCCRQ/P capability advertisement. That seems inline with
existing flow, and solves the requirement. Would that work?

Regards,

--Carlos (as editor).

>  
> Regards
> yp
>> Thoughts?
>>
>> Thanks,
>>
>> --Carlos.
>>
>>
>>> Regards
>>> yp
>>>
>>>> May be we can have this rule?
>>>>
>>>> - If receiver of this AVP decides to support this feature he should
>>> ack
>>>> the same so that sender knows for sure he can expect alignment that
> he
>>>> needs.
>>> Note also that both directions could request a different offset
> size,
>>> much like requesting (or not) and L2SS, or requesting (or not)
> different
>>> Cookie size (and value).
>>>
>>>> - Else sending this AVP is a nop to sender/receiver.
>>>>
>>> I think that per-tunnel capability and per-session request is
> simpler,
>>> and (more importantly) aligned with how other AVPs/capabilities are
>>> currently signaled: Tunnel-level capabilities (from example from
> RFC3573
>>> that Ignacio cited, and all the "xxx Capabilities" AVPs) and at the
>>> session-level, the downstream (receiver, decap) requests what he
> wants
>>> sent to it (L2SS, Cookie, Session ID, etc).
>>>
>>>> My $.02 :), of course.
>>> Much appreciated. What do you think?
>>>
>>> Thanks,
>>>
>>> --Carlos.
>>>
>>>> Regards
>>>> yp
>>>>
>>>>
>>>> The "Offset Capability" AVP can have the same syntax as the "Offset
>>>> Size" with different semantics. It could even be the same AVP
> number
>>>> (with semantics depending on the type of message in which the AVP
> is
>>>> in), or (my preference) a new AVP.
>>>>
>>>> That would allow the sender to have control via capability, and
> would
>>>> default to #1 if the "Offset Capability" is missing or it has a
> zero
>>>> value. Would that be an acceptable solution?
>>>>
>>>> Thanks,
>>>>
>>>> --Carlos (as editor).
>>>>
>>>>> The received value of this new AVP, together with the sent Offset
>>>>> Size AVP will tell the receiver what to expect in terms of the
>>>>> offset padding field: if the sender said it is willing to insert
>>>>> up to 4 bytes and the receiver asked (or was going to ask) for 6,
>>>>> only 4 bytes will be inserted by the sender. If the sender said up
>>>>> to 8 bytes and the receiver asked for 6, the sender will insert
>>>>> 6 bytes. If the sender said no offset will be inserted, nothing
>>>>> will be inserted.
>>>>>
>>>>> Again, my preference is #1 as that would eliminate a significant
>>>>> protocol complexity that adds no benefit (IMHO).
>>>>>
>>>>> Opinions, please?
>>>>>
>>>>> -Ignacio
>>>>>
>>>>> PS: This draft expired recently but a new one is in the works.
>>>>> In the meantime, you can look here for the last version:
>>>>> http://tools.ietf.org/html/draft-ietf-l2tpext-l2tp-ppp-05
>>>>>
>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> L2tpext mailing list
>>>>> [e-mail]
>>>>> https://www1.ietf.org/mailman/listinfo/l2tpext
>>>>>
>> --
>> --Carlos Pignataro.
>> Escalation RTP - cisco Systems
> 

-- 
--Carlos Pignataro.
Escalation RTP - cisco Systems


_______________________________________________
L2tpext mailing list
[e-mail]
https://www1.ietf.org/mailman/listinfo/l2tpext