The MPLS WG Archive

Cell Relay Retreat>MPLS WG Archive>month:2001-Feb> msg00316



[Date Prev][Date Next][Thread Prev][Thread Next]  
  [Date Index][Thread Index][Author Index][Subject Index]

New TTL Draft

  • From: Puneet Agarwal <puneet@pluris.com>
  • Date: Mon, 26 Feb 2001 15:00:17 -0800
  • Cc: "'mpls@uu.net'" <mpls@UU.NET>, "'flefauch@cisco.com'" <flefauch@cisco.com>, "'liwwu@cisco.com'" <liwwu@cisco.com>, "'bsd@cisco.com'" <bsd@cisco.com>, Bora Akyol <akyol@pluris.com>

Hi Shahram,

Thanks for your comments. My comments are inline.

-Puneet

>
>Hi Puneet/Eric,
>
>1)You are right we have not described how to signal Pipe/short 
>pipe/uniform models. Since our draft is in its last call and 
>can't be changed substantially, I propose that we get together 
>to write a separate I-D to address the CR-LDP/RSVP-TE 
>extensions for signaling the tunneling model.

Agreed. We can take this up privately with each other.

>
>2) Section 2.4 does not discuss an important case that is: 
>egress processing of tunnels, i.e. pop and forwarding the 
>packet as MPLS. 

Pop (or Egress Pop to distinguish from PHP) is actually covered in section
2.3. The possibly operation when forwarding the packet as MPLS are covered
in the 3 bullets  of section 2.4. Combine section 2.3 and the relevant
section of 2.4 and I think your concern about the missing case should be
answered. Hence as far as I can see the case is convered, but please let me
know if you still think that it is not addressed.
>
>3)As you may know, version -08 of our draft has introduced two 
>types of Pipe tunnels a) Short pipe and b) Pipe. The short 
>pipe is what was called the "pipe" in version -07. But the new 
>"pipe" model is a bit different, in the sense that all the 
>egress processing is done based on top label. So your 
>description of TTL processing for the pipe model remains valid 
>for "short pipe" (except that you need to describe the egress 
>processing too, see bullet 2 above), but you need to define 
>another TTL processing rule for the "pipe", which is:  the top 
>label TTL is decremented and checked, while the lower label 
>TTL (unchanged) is used for forwarding the packet. 

We had based our draft on -07 version of the mpls-diff draft. I will take
into account the -08 version (where the new model nomenclature was
introduced) and update our spec accordingly.

-Puneet
>
>Yours,
>-Shahram
>
>
>
>
>> -----Original Message-----
>> From: Puneet Agarwal [mailto:puneet@pluris.com]
>> Sent: Friday, February 23, 2001 4:53 PM
>> To: 'Eric Gray'
>> Cc: 'mpls@uu.net'; 'flefauch@cisco.com'; 'liwwu@cisco.com';
>> 'bsd@cisco.com'; Shahram Davari; Bora Akyol
>> Subject: RE: New TTL Draft
>> 
>> 
>> Eric,
>> 
>> Comments inline.
>> 
>> >-----Original Message-----
>> >From: Eric Gray [mailto:eric.gray@sandburst.com]
>> >Sent: Friday, February 23, 2001 9:34 AM
>> >To: Bora Akyol
>> >Cc: akyol@akyol.org; djiang1@lucent.com; B, Prakash K (Prakash);
>> >'mpls@uu.net'
>> >Subject: Re: New TTL Draft
>> >
>> >
>> >Bora/Puneet,
>> >
>> >    Some comments.
>> >
>> >    In section 2.4, item 3 - PHP does not necessarily expose an
>> >IP header.  The text in this section is not clear in this respect
>> >- possibly because of the textual construct "IP-Header/label".
>> >It might be somewhat clearer if you use "header" and add a
>> >paragraph stating that the PHP exposed header may be either
>> >an IP header or an MPLS label and the appropriate oTTL  is
>> >written to the corresponding TTL field of this PHP exposed
>> >header.
>> 
>> I agree. We will change the wording in the next rev of the draft. 
>> >
>> >    Also, in this section, why is oTTL set to the TTL of the
>> >exposed header if it is not used for anything?
>> 
>> Good point. We do not need to set this. This was a leftover 
>> from some text
>> we took out of section 2.5 at the last moment that used this 
>> information. We
>> will fix this in the next rev.
>> >
>> >    In addition, if your actually specifying a behavior (rather than
>> >an architecture), I believe you need to allow that the TTL may
>> >be decremented by a number other than one.  RFC 791 states
>> >that RFC is decremented by at least one.  Your draft states in
>> >several places that iTTL is decremented (I assume that the
>> >underscore in these cases is meant to be a minus sign) by one.
>> 
>> The underscore is indeed supposed to be a minus. Will be 
>> fixed in the next
>> rev.
>> I have no problems in allowing the iTTL to be decremented by 
>> more than one.
>> I will change the text to state that the iTTL should be 
>> decremented by "N",
>> where "N" is user configured and is greater than 0. The 
>> default value of "N"
>> should be 1.
>> 
>> >
>> >    There are practical reasons for decrementing TTL by more
>> >than one (a high-cost or low-bandwidth link, for example)
>> >when it is desirable to accelerate the aging of possibly looping
>> >packets.  A higher TTL decrement punishes all packets that use
>> >the link, but it punishes looping packets much more.
>> >
>> >    Finally, what is the mechanism which the PHP uses to determine
>> >whether or not the LSP is using the pipe or the uniform model?  I am
>> >not sure this information is currently available.
>> 
>> You may be right. I looked through 
>> "draft-ietf-mpls-diff-ext-08.txt", but
>> did not see any method to signal the model (pipe or uniform) 
>> associated with
>> the LSP, using either RSVP or LDP. If it is a missing piece, 
>> then I hope
>> that it will be addressed by the authors of that draft.
>> 
>> -Puneet
>> 
>> >
>> >--
>> >Eric Gray
>> >
>> >Bora Akyol wrote:
>> >
>> >> Note that we (Puneet and I) just posted an Internet Draft on TTL
>> >> handling in hierarchical MPLS networks.
>> >>
>> >> The internet draft is titled: draft-agarwal-mpls-ttl-00.txt
>> >>
>> >> 
>> http://search.ietf.org/internet-drafts/draft-agarwal-mpls-ttl-00.txt
>> >>
>> >> We welcome any comments,
>> >>
>> >> Bora Akyol
>> >
>> 
>