Shahram:

Thanks for your quick reply.  Please see my new comments inline.

Rich

-----Original Message-----
From: Shahram Davari [mailto:[email protected]] 
Sent: Wednesday, October 24, 2012 10:39 AM
To: Richard Tse; [email protected]
Cc: [email protected]
Subject: Re: [TICTOC] I-D Action: draft-ietf-tictoc-1588overmpls-03.txt

Hi Richard,

Thanks for your comments. Please see my responses inline.

Regards,
Shahram

-----Original Message-----
From: Richard Tse [mailto:[email protected]] 
Sent: Tuesday, October 23, 2012 8:27 PM
To: [email protected]
Cc: [email protected]; Richard Tse
Subject: RE: [TICTOC] I-D Action: draft-ietf-tictoc-1588overmpls-03.txt

Shahram:

I have a few comments about this draft:

A)  In clause 3, you say "...the payload can't be deterministically identified 
by LSRs because the payload type is a context of the PW label and the PW label, 
and the PW label and its context are only known to the Edge routers (PEs/LERs); 
LSRs don't know what is a PW's payload (Ethernet, ATM, FR, CES, etc)."

However, a new contradictory text in clause 7 says:  "It is the job of the 
LER/LSR to parse the timing message and find out the type of the Timing message 
and decide whether and how to Time-stamp it (e.g. , BC) or modify existing 
timestamp in it (e.g., TC)."  

I think the new text in clause 7 needs to be changed.  I don't think we want 
LERs and LSRs to look inside timing packets to decide how to timestamp it.  I 
think the timing LSPs should be unique and should contain all the information 
that the LER/LSR needs for its timestamping function.  

SD> the two statements you mentioned are not contradictory. Based on MPLS 
architecture the payload of PW is only known to the ingress and Egress LERs and 
not to intermediate LSRs.  This draft proposed to signal whether the payload 
for a specific Timing LSP is  IP or PW so that all LSRs in the path know the 
payload. The signaling also specifies  what type of timing message is 
encapsulated (PTP, NTP, etc). So a parser can easily parse the Timing message. 
Parsing of Timing message is required since not all Timing messages require 
Time-stamping or TC correction.  For example 1588 has event messages and 
non-event messages. 1588 standard requires different time-stamping behavior for 
different messages.

[Richard Tse] I see.  From the text in clause 3, my original interpretation was 
that different labels would be used for different message types.  Thus, the 
LSRs and LERs would not have to look at anything except the label.  I really 
liked this because it was so simple and had only the minor(?) cost of using a 
few more labels.    But, I guess this would be corrupted by the OAM and Mgmt 
messages which also use the PTP LSPs and require different actions.  



B). Following along the lines of my comment above regarding deep packet 
inspection, how is 2-step functionality handled in a transparent clock in the 
MPLS network?  To update a follow_up packet, an LSR TC would need to look 
inside the packet to associate the sequenceID of the Follow_up message with 
that of the Sync message.  Perhaps you should allow only one-step timestamping 
in all MPLS nodes.  Follow_up packets that were generated outside the MPLS 
network can still pass through, but they will not be modified.  This would 
simplify the timestamping function and negate the need for any deep-packet 
inspection.

[SD] For 2-step the CPU can do the parsing and deep packet inspection is not 
required at line rate in HW. 

[Richard Tse]  I was hoping to simplify the implementation so the LERs would 
not need a deep-packet processing element to handle transparent clocks.  With 
2-step clocks, the CPU needs to hold a lot of PTP context to wait and associate 
follow_up packets with sync packets.  When doing deep-packet inspection and 
2-step clock context handling, nominal PTP packet rates multiplied by the many 
PTP connections that may flow through a TC will require non-insignificant 
processing power.  With 1-step clocking, this could all be done in H/W.  



C).  Clause 6.4 of the newly released draft of ITU-T G.8275.1, Precision time 
protocol telecom profile for phase/time synchronization, states that "The 
Ethernet mapping, as per IEEE 1588-2008 Annex F, is agreed as the default 
mapping for the profile."  Sooner or later, you will need to add the 
PTP-over-Ethernet-over-MPLS encapsulation to section 6 of your standard to 
support this profile.

[SD] The method for transporting Ethernet over MPLS is via PW and that is 
covered in section 6.2. 
[Richard Tse] Thanks.  I overlooked the fact that you removed all the IP and 
UDP layers from the PW encapsulation. 


----------


Richard Tse
Technical Advisor
PMC-Sierra, Inc
8555 Baxter Place
Burnaby, BC
Canada
V5A 4V7
Phone:  +1-604-415-6015
Email:  [email protected]
www.pmc-sierra.com


-----Original Message-----
From: [email protected] [mailto:[email protected]] 
Sent: Monday, October 22, 2012 1:28 AM
To: [email protected]
Cc: [email protected]
Subject: [TICTOC] I-D Action: draft-ietf-tictoc-1588overmpls-03.txt


A New Internet-Draft is available from the on-line Internet-Drafts directories.
 This draft is a work item of the Timing over IP Connection and Transfer of 
Clock Working Group of the IETF.

        Title           : Transporting Timing messages over MPLS Networks
        Author(s)       : Shahram Davari
                          Amit Oren
                          Manav Bhatia
                          Peter Roberts
                          Laurent Montini
        Filename        : draft-ietf-tictoc-1588overmpls-03.txt
        Pages           : 36
        Date            : 2012-10-22

Abstract:
   This document defines the method for transporting Timing messages
   such as PTP and NTP over an MPLS network.  The method allows for the
   easy identification of these PDUs at the port level to allow for port
   level processing of these PDUs in both LERs and LSRs.

   The basic idea is to transport Timing messages inside dedicated MPLS
   LSPs.  These LSPs only carry timing messages and possibly Control and
   Management packets, but they do not carry customer traffic.

   Two methods for transporting Timing messages over MPLS are defined.
   The first method is to transport Timing messages directly over the
   dedicated MPLS LSP via UDP/IP encapsulation, which is suitable for
   MPLS networks.  The second method is to transport Timing messages
   inside a PW via Ethernet encapsulation, which is suitable for both
   MPLS and MPLS-TP networks.


The IETF datatracker status page for this draft is:
https://datatracker.ietf.org/doc/draft-ietf-tictoc-1588overmpls

There's also a htmlized version available at:
http://tools.ietf.org/html/draft-ietf-tictoc-1588overmpls-03

A diff from the previous version is available at:
http://www.ietf.org/rfcdiff?url2=draft-ietf-tictoc-1588overmpls-03


Internet-Drafts are also available by anonymous FTP at:
ftp://ftp.ietf.org/internet-drafts/


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

Reply via email to