Cyril, 

Thank you for the review, please see inline below [ina]

Ina

-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Margaria, 
Cyril (Coriant - DE/Munich)
Sent: Wednesday, October 09, 2013 6:55 AM
To: [email protected]
Cc: [email protected]
Subject: Re: [Pce] I-D Action: draft-ietf-pce-stateful-pce-07.txt

Hi, 

I have a some comments on this revision: 

1) Section 3.1 seems quite generic and mainly related to Stateful PCE and not 
the extensions themselves, as the document is quite big (46 pages), it could be 
considered to shift the text to the applicability draft.

[ina] The draft has been significantly slimmed down, from 59 to 46 pages:-) 
Given this is a base document, a generic section providing some background is 
useful and I prefer to leave it in this draft.
 

2) One important change is that the Stateful PCE does NOT allow anymore to 
reference the LSP in the PCReq and PCRep messages. This make the stateful PCEP 
extensions a pure LSP-DB update mechanism, this is quite a change that drops  a 
set of useful Stateful functionalities (routing policy based on LSP sender, 
planned path, ...etc). In addition this do es remove the capability of a PCE to 
re-order the PCRep mentioned in section 3.1.2.

[ina] We don't have the corresponding use cases in the applicability document 
for the pcreq/pcrep extensions using the LSP object. I believe the gmpls 
document does cover such use cases for GMPLS.

It would help if we could have a summary of the changes (and why), I find the 
path of having a Path Computation Element Protocol stateful extension not 
allowing LSP state in path computation request. 



3) During discussion some important architectural consideration were mentioned, 
but they are not described in section 5 In Section 5.2 the reason indicated for 
having in-band PCRpt and PCUpd was to simplify the PCE operation for one PCC by 
serializing the Requests and the update message for easier correlation of 
events. Furthermore the document requires this (which I believe can be worked 
around by never sending anything if an existing LSP-DB update protocol is used) 

[ina] Using an LSP-DB update protocol is not a mode of operation we are 
supporting with this draft. 

4) Section 5.3

The capability advertisement reflects a design principle, which is not stated: 
the extensions requires the support of PCRpt message for the reason that should 
be addressed by 3).
The drawback of this mandated PCRpt support is at minimum a RECOMMENDED 
extended session opening mechanism. I believe the reason is to remove the 
number of supported options. If this is the case (and if there are other 
reasons), it would be good if this is stated. 
 
[ina] Yes, we don't support the model where state is communicated some other 
way. 

5) Section 5.3

   Editorial : "Note that even if the
   update capability has not been advertised, a PCE can still receive
   LSP Status Reports from a PCC and build and maintain an up to date
   view of the state of the PCC's LSPs."
Could be removed

[ina] I think you are thinking of a model where the state is communicated some 
other way, but this is not what we are supporting in this draft. 

6) Section 5
The section title is "Architectural Overview of Protocol Extensions", yet it 
describes procedures. I think that the content should be split into 
architectural aspects and procedures.  For example. RFC5440 does have an 
architecture section, which summarize the procedures, but the procedures are 
described in each message. 
For instance section 5.4 describe session opening extension, a separate section 
should describe the procedure, separated from architectural principle. 
Procedure are described starting at paragraph 4. This apply to the other 
sub-sections too.

[ina] The section describes the operation at high level. We can rename it to 
reflect that. 

7) Section 6.1
OLD
   "If the PCRpt message is in response to a PCUpd message, the SRP
   object SHOULD be included and the value of the SRP-ID-number in the
   SRP Object MUST be the same as that sent in the PCUpd message that
   triggered the state that is reported." 
NEW
   "If the PCRpt message is in response to a PCUpd message of the same PCEP 
session,
   the SRP object MUST be included and the value of the SRP-ID-number in the
   SRP Object MUST be the same as that sent in the PCUpd message that
   triggered the state that is reported." 

[ina] It is SHOULD, not MUST, because there is no way to enforce it (no error 
can be defined to catch this case). 

If the object SHOULD be present, this is a pure recommendation, but section 7.2 
(SRP object) indicates that each id is considered unacknowledged (and should 
not be reused) if no PCRpt comes with an higher id (for that LSP). If the 
recommendation is not followed all ids may end up unacknowledged.
The second point is that is must be scoped to the session that sent the PCUpd, 
(the D bit also?) otherwise PCEs NOT having the delegation will interpret that 
as delegation request, which it may allow, then the behavior is up to the PCC 
implementation (One policy could be to delegate to the newest in case the TCP 
session is hanging).

8) Section 6.1 

You forbid state compression on PCRpt, only on PCUpd. I do not see the benefit, 
could you detail why? 
[ina] To allow correlation or errors and requests.

Soft state signaling protocol may introduce some spurious state change. 
[ina] This is ok, they will be reported with a 0 srp-id-number. 

The document states no FSM for the states, so it should not matter. 


9) Section 6.2
   Regarding the Make-before-break : there will be 2 path co-existing, they 
should have then different symbolic-path-name, correct?
[ina] No, the symbolic name is equivalent to the lsp name in the configuration. 

10) Section 6.2 
   PCUpd state compression and SRP-ID : its allowed to compress updates, 
keeping the most recent one. A provision should be made in case of SRP-ID 
wrap-up, otherwise high SRP-ID will not be acknowledged.
[ina] Good point, thank you. 

11) section 6.3
 - RFC5511 does not define comments, 
 - The format makes me wonder why the RP is not used instead of the SRP. I 
understand that the id-space are different, but a new bit in the RP flag could 
easily distinguish it (PCE RP (1) versus PCC RP (0)) what would be the 
drawbacks of it? The advantage would be that we do not need to redefine the 
PCErr and reuse the existing definitions and mechanisms for RP.     
[ina] We evaluated using the RP and decided against it. 

12) Section 7
 P and I are note related to path computation requests only, I do not see why 
its optional, the P & I Bit would be very useful in PCUpd messages.
[ina] Sorry, can you rephrase, I didn't understand the comment. 

13) Section 7.1.1
  A PCEP document must not mandate the use of a particular signaling protocol, 
the RSVP part must be removed. 
[ina] I agree with you in principle. However, the reality is that current use 
of pcep implies RSVP signaling. This was not a problem in the past, but now 
that SR extensions are also possible, it makes sense to state this. Do you have 
a proposal how to reword?

14) section 7.2
  Why a state pending is not an ack? This seems to mix the PCEP protocol 
mechanism and the provisioning mechanism. Could you explain the reason for this 
choice (at least per mail)
[ina] Because we wanted to ensure ack means that the upd was acted on. 

15) Section 7.2 
        Why can the SRP contains a path name when its already present in the 
mandatory LSP object? 
[ina] Because you may generate an error that does not contain the LSP object 
(e.g. the lsp was not created). 


16) Section 7.2
 As the ID is present in PCRpt, which may be sent to different session after an 
PCUpd, a reference to section 6.1 on the mirroring of the SRP per session or a 
note could be useful 
[ina] Sorry, don't understand the comment, can you rephrase?

17) Section 7.3
 D bit should be scoped to the PCEP session having the delegation. If this is 
correct please it.
S bit should also be scoped to the PCEP session, this is kind of obvious but it 
does not hurt to state it. 
[ina] Sorry, not sure I follow, can you rephrase. 


18) Section 7.3 
  This raise the point of how can you identify an LSP, it seems that one LSP 
can have different RSVP identifiers (as you state). Could you care to explain 
how to identify an LSP in that case, because I am confused.
[ina] The only identifier is the plsp-id, all the other identifiers you can 
think of as attributes of the LSP. More below. 

19) Section 7.3.1
 I would remove the RSVP-signaled statement. 
 From protocol mechanism there are  the following LSP identifiers, with 
different relation : 
  PLSP-ID            : Unique per PCEP session, mandatory, represent a LSP
  LSP-Identifier TLV : mandatory 
  SYMBOLIC-PATH-NAME : mandatory for a path and LSP 
An LSP may have different LSP-Identifier during its lifetime, (what happens if 
the same LSP-Identifier appear with a different PLSP-ID?)

[ina] the lsp identifier is tightly tied to the RSVP signaling mechanism, it 
will not apply for SR-related lsps for example. The symbolic-name is the human 
identifier, equivalent to the configured name in implementations today. It 
could have been used instead of plsp-id but that would have been cumbersome. 
  
You indicate that a LSP is a path (section 7.3.2), there is a 1:1 relationship 
between a PLSP-ID and a path name, but potentially an arbitrary relationship 
between PLSP-ID and LSP-Identifiers. 
This is quite confusing, I do not understand why so many identifiers are 
required. 
[ina] the lsp identifier is the rsvp session information, and that changes 
(e.g. after reoptimization). Please note that the lsp identifiers is tightly 
tied to RSVP (to your point previously about the signaling protocol)

I would understand an unique symbolic LSP id (I dislike the term name, it can 
be any identifier and not a string) that is mapped per session to PLSP-ID for 
efficiency. 
In addition to this PCC unique identifier the LSP identifier TLV (optional) or 
LSP name (ascii string, optional) could be added. 

What would speak against this proposal?  

In addition there should be only one symbolic LSP id TLV per LSP object.
[ina] Can you explain how this proposal will work for mbb for RSVP?

20) section 7.3.3 and section 7.3.4

There is 2 TLV for the error reporting, presence rule are more complicated.

      0                   1                   2                   3
      0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
     |           Type=[TBD]          |            Length=variable    |
     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
     |                          LSP Error Code                       |
     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
     | Additional Error type         |  Additional error length      |  
     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
     |                  Additional Error information                 |
     ~                   Depend on LSP Error code                    ~
     |                                                               |
     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
        

The additional error type can be RSVP or String , when RSVP the format of 7.3.4 
is used.

[ina] Are you proposing a new error way to encode errors? What are the issues 
with the current error encoding? 

Addition question : how many TLVs can be present in the LSP object? 

[ina] Sorry, not sure where you are going with this, how is this different than 
other objects with a variable number of TLVs?


Mit freundlichen Grüßen / Best Regards
Cyril Margaria

> -----Original Message-----
> From: [email protected] [mailto:[email protected]] On Behalf Of
> [email protected]
> Sent: Wednesday, October 09, 2013 8:42 AM
> To: [email protected]
> Cc: [email protected]
> Subject: [Pce] I-D Action: draft-ietf-pce-stateful-pce-07.txt
> 
> 
> A New Internet-Draft is available from the on-line Internet-Drafts
> directories.
>  This draft is a work item of the Path Computation Element Working Group of
> the IETF.
> 
>       Title           : PCEP Extensions for Stateful PCE
>       Author(s)       : Edward Crabbe
>                           Jan Medved
>                           Ina Minei
>                           Robert Varga
>       Filename        : draft-ietf-pce-stateful-pce-07.txt
>       Pages           : 47
>       Date            : 2013-10-08
> 
> Abstract:
>    The Path Computation Element Communication Protocol (PCEP) provides
>    mechanisms for Path Computation Elements (PCEs) to perform path
>    computations in response to Path Computation Clients (PCCs) requests.
> 
>    Although PCEP explicitly makes no assumptions regarding the
>    information available to the PCE, it also makes no provisions for
>    synchronization or PCE control of timing and sequence of path
>    computations within and across PCEP sessions.  This document
>    describes a set of extensions to PCEP to enable stateful control of
>    MPLS-TE and GMPLS LSPs via PCEP.
> 
> 
> 
> The IETF datatracker status page for this draft is:
> https://datatracker.ietf.org/doc/draft-ietf-pce-stateful-pce
> 
> There's also a htmlized version available at:
> http://tools.ietf.org/html/draft-ietf-pce-stateful-pce-07
> 
> A diff from the previous version is available at:
> http://www.ietf.org/rfcdiff?url2=draft-ietf-pce-stateful-pce-07
> 
> 
> Please note that it may take a couple of minutes from the time of submission
> until the htmlized version and diff are available at tools.ietf.org.
> 
> Internet-Drafts are also available by anonymous FTP at:
> ftp://ftp.ietf.org/internet-drafts/
> 
> _______________________________________________
> Pce mailing list
> [email protected]
> https://www.ietf.org/mailman/listinfo/pce


_______________________________________________
Pce mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/pce



_______________________________________________
Pce mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/pce

Reply via email to