RE: [T1X1.5] Re: LCAS and draft-mannie-ccamp-gmpls-lbm-tdm-00.txt

[Steve West <[e-mail]>]

Eric:

Thanks for your reply.

Please see my comments below.

Kind regards,

Steve

>    -----Original Message-----
>    From: Mannie, Eric [mailto:[e-mail]]
>    Sent: Monday, November 19, 2001 2:28 AM
>    To: 'Dimitri Papadimitriou'; Steve West
>    Cc: 'Stephen Trowbridge'; 'Maarten Vissers'; ccamp; q11/15; t1x1.5
>    Subject: RE: [T1X1.5] Re: LCAS and
>    draft-mannie-ccamp-gmpls-lbm-tdm-00.txt
>    
>    
>    Hello Steve,
>    
>    Just to complement the answer of Dimitri,
>    
>    > You believe that the packet loss rate and robustness of LMB is
>    > similar to LCAS ?
>    
>    LCAS waits for the end of control packet which is 
>    multi-framed to decide
>    when the payload is significant (the "transition"). LBM 
>    uses for instance
>    the C2 byte for the HOVC, since this byte appears directly 
>    in the POH and is
>    not part of a control packet, that should go faster, right 
>    ? I.e. less
>    glitch than with LCAS ?

To avoid data corruption, the two ends of the link must agree precisely when
payload data is added/deleted.  This unavoidably requires hardware level
synchronization.

Delay does not necessarily imply data corruption.   Consider for example the
case where the Virtual Concatenated bandwidth capacity is being added before
turning up the new services.

Note also that centralized service management with parallel provisioning of
all nodes on the path and  end-to-end coordination through LCAS H4 may
actually have lower delay than hop-by-hop coordination through
RSVP-TE/GMPLS. 

>    
>    But having a field in the transport plane to indicate a 
>    transition (whatever
>    is the field ) is not the most important issue. Even if 
>    you have that field,
>    you need some time to provision the additional VCs/SPEs in 
>    the network. This
>    is not done by LCAS, and any control plane or management 
>    plane will imply
>    some time for that operation, and the glitch will really 
>    come from that
>    process, not from the way that you detect a transition in 
>    the transport
>    plane. 

To support true hitless transitions in the transport plane does require
bit-accurate coordination by the mapper hardware at the two ends.

>    
>    > 4.  LBM is an extension to GMPLS.  GMPLS is required for 
>    LBM, whereas LCAS
>    > could potentially be used with GMPLS or with other 
>    control plane or
>    > management plane mechanisms ?
>    
>    When a distributed control plane is used (e.g. GMPLS or 
>    PNNI), using LCAS
>    implies issues on this control plane. LCAS is a 
>    stand-alone solution that
>    bypass the control plane to change the resource 
>    reservation in both end
>    systems. A control plane maintains states in end systems 
>    (e.g. at the UNI)
>    and in intermediate systems ! So, bandwidth modifications 
>    must be reflected
>    in the control plane otherwise it will become inconsistent.  
>    
>    Now, the control plane is natively designed to deal with 
>    these resource
>    reservations including in the source and the destination 
>    (with the UNI).
>    Moreover, GMPLS is natively designed to deal with 
>    bandwidth modification (in
>    the packet world this was one of the major requirements). 
>    Originally, we
>    just wanted to describe how to generalize these bandwidth 
>    modification
>    principles to SDH/SONET and how to interwork with LCAS.
>    

Agreed.  Some management or signaling mechanism will be required to setup
cross connects at every node the path traverses.  However this is
independent of the need to synchronize the payload mappers.


>    Then, we saw that a large part of LCAS was overlapping 
>    with GMPLS, i.e. all
>    the protocol part. The rest was the issue about when to 
>    decide that a VC/SPE
>    contains real user data or not (the transition).
>    

The LCAS state machines are a small subset of what is required to support
the end to end connection.


>    Since LCAS added a control message, naturally it used a 
>    field in that
>    control message to indicate that transition. But with 
>    GMPLS we didn't need
>    any new control message, se we looked at what was available *in the
>    transport plane* to detect the fact that a VC/SPE contains 
>    now real user
>    data (i.e the transition). We found natural to use the 
>    so-called SDH/SONET
>    "signal label" (e.g. the C2 byte) (note for the readers: 
>    nothing to do with
>    the GMPLS label :-). So we added these aspects to the 
>    draft. This is the
>    part that we expected to validate with the ITU-T/T1X1 
>    folks, but flames came
>    first...
>    

IMHO  H4 is a better choice because it is already used to carry the Virtual
concatenation Sequence number, and member additions / deletions require
informing the mapper which sequence numbers are valid.


>    What we are certainly going to do, is to add in the draft 
>    a section about
>    how GMPLS should work when LCAS is used at the same time. 

Great!

>    However, we think
>    that using GMPLS and LCAS at the same time will be more 
>    complex than simply
>    using GMPLS and will not bring something additional. So the natural
>    observation came, "why should we use LCAS with GMPLS and 
>    build something
>    more complex when there is almost no issue to solve when 
>    simply using GMPLS
>    alone". Using GMPLS and LCAS at the same time to modify 
>    the bandwidth
>    implies that we will have two signaling flows between the 
>    source and the
>    destination running in parallel and overlapping. GMPLS 
>    will also speak with
>    the source and the destination to do the same as LCAS, and with a
>    distributed control plane you need GMPLS anyway (or PNNI or etc).
>    
>    We also added a section to explain how several GMPLS LSPs 
>    (i.e. circuits)
>    can be combined together to form a larger virtually 
>    concatenated circuit.
>    This allows to deal with the non co-routing of VCs/SPEs with GMPLS.

I agree that this is valuable.  

Also, in some applications (e.g. Internet ISP backbones), absolutely hitless
operation may not be necessary, so your make-before-break proposal would
work. 

But for traffic engineered virtual private line services with strict SLAs,
absolutely hitless operation is necessary.  And LCAS is well designed for
this purpose.

>    
>    Kind regards,
>    
>    Eric
>    
>    -----Original Message-----
>    From: Dimitri Papadimitriou 
>    [mailto:[e-mail]]
>    Sent: Sunday, November 18, 2001 3:44 PM
>    To: Steve West
>    Cc: Mannie, Eric; 'Stephen Trowbridge'; 'Maarten Vissers'; ccamp;
>    q11/15; t1x1.5
>    Subject: Re: [T1X1.5] Re: LCAS and
>    draft-mannie-ccamp-gmpls-lbm-tdm-00.txt
>    
>    
>    Hi Steve,
>    
>    Thanks for your comments, see in-line...
>    
>    Steve West wrote:
>    > 
>    > Dimitri:
>    > 
>    > Please confirm my understanding of your proposal:
>    > 
>    > 1.  Your proposal supports hitless link capacity 
>    adjustment. (Capacity
>    > changes that occurs during transmission of a packet will 
>    not corrupt that
>    > packet.)  You believe that the packet loss rate and 
>    robustness of LMB is
>    > similar to LCAS ?
>    
>    It is mentioned in the proposal that the source waits for a 
>    confirmation from the destination (i.e. Resv Message). THEN 
>    the source "equip" the VC/SPE and sends. After the destination 
>    sees in the POH that the VC is "equipped" (transition).
>    
>    So instead of using the transitions proposed in LCAS (based 
>    on the H4 CTRL words) we use the ones already available.
>    
>    Since we are the beginning, we were very cautious but we expect to 
>    reach more than reasonable robustness and reliability.
>    
>    > 2.  LMB requires hardware coordination between the SONET 
>    payload mapper
>    and
>    > the SONET POH processor (C2 octet insertion/extraction logic) ?
>    
>    At the transport plane level C2 is the field that LBM uses for the 
>    transition in the HOVC POH.
>     
>    > 3.  The hardware based synchronization performance 
>    requirements are
>    similar
>    > to those required by LCAS.  Except that LCAS coordinates 
>    the critical
>    > transitions using H4 instead of C2 ?  And that Virtual 
>    Concatenation
>    systems
>    > require to coordinate their payload mappers with bits 
>    1-4 of the H4 byte
>    to
>    > determine the sequence number anyway ?
>    
>    See above, yes and yes (since we re-use existing VC mechanisms)
>     
>    > 4.  LBM is an extension to GMPLS.  GMPLS is required for 
>    LBM, whereas LCAS
>    > could potentially be used with GMPLS or with other 
>    control plane or
>    > management plane mechanisms ?
>    
>    The scope of this proposal is to use GMPLS-LBM without 
>    LCAS. In fact 
>    (from what i understand in your question) see it more from 
>    transport 
>    to control plane impact taking the assumption that you expect some 
>    hardware improvement. Let's first determine if this can be 
>    achieved 
>    and then discuss the implementation issues that can arise 
>    from this 
>    proposal. In a sense, GMPLS-LBM is a self-consistent solution.
>    
>    Regards,
>    - dimitri.
>     
>    > Thanks,
>    > 
>    > Steve
>    > 
>    > -----Original Message-----
>    > From: Dimitri Papadimitriou 
>    [mailto:[e-mail]]
>    > Sent: Friday, November 16, 2001 4:09 PM
>    > To: Mannie, Eric; 'Stephen Trowbridge'; 'Maarten Vissers'
>    > Cc: ccamp; q11/15; t1x1.5
>    > Subject: [T1X1.5] Re: LCAS and 
>    draft-mannie-ccamp-gmpls-lbm-tdm-00.txt
>    > 
>    > Hi folks,
>    > 
>    > What happens... when we start the technical discussion
>    > all guys taking part to the previous one now are gone ?
>    > 
>    > We would like to start a technical discussion.. rather
>    > difficult if we stay alone... or everybody agree on
>    > proposal made in the previous e-mail ? (this in order
>    > to refine the mechanism detailed in the GMPLS LBM I-D.)
>    > 
>    > Any technical feedback is highly appreciated.
>    > 
>    > Cheers,
>    > - dimitri.
>    > 
>    > "Mannie, Eric" wrote:
>    > >
>    > > Hello Stephen
>    > >
>    > > > GMPLS-LBM and LCAS ARE solutions to different problems.
>    > >
>    > > No, please see explanations below.
>    > >
>    > > > If you use LBM and change the mappings at the two 
>    ends without some
>    > > coordination such as LCAS, you will have a glitch to 
>    the traffic
>    > > (I think this is simple enough).
>    > >
>    > > No, please see below (C2 or V5 transition).
>    > >
>    > > >The compare/contrast discussion of LBM and LCAS is 
>    appropriate not
>    > > >because LCAS is "sacred", but because you are 
>    comparing apples to
>    > > >oranges. Statements in this document such as "One 
>    drawback of LCAS is
>    .."
>    > > >are unnecessary and inappropriate in this context, 
>    since LBM does
>    > > >NOT solve the problem LCAS was designed to solve. 
>    Perhaps if you
>    > > >were solving the SAME problem better, this would be 
>    OK, but this
>    > > >is not the case.
>    > >
>    > > No, please see below.
>    > >
>    > > Simple example:
>    > >
>    > > 1) you first signal an additional VC/SPE to the 
>    destination (via a
>    Path).
>    > > 2) the destination knows exactly which VC/SPE is added 
>    (via the label).
>    > > 3) the new VC is unequipped (indicated in POH C2 (for HOVC)),
>    > >    so the VC/SPE is not yet added to concatenated signal.
>    > > 4) then, you wait at the source for a confirmation 
>    from the destination
>    > > (i.e. Resv).
>    > > 5) the source "equip" the VC/SPE and sends.
>    > > 6) the destination sees in the POH that the VC is "equipped"
>    (transition).
>    > >
>    > > The difference between LCAS and LBM is that LCAS uses 
>    an additional
>    > explicit
>    > > indication in the overhead to say "this is the 
>    payload" to use. LBM
>    > doesn't
>    > > change the SDH/SONET overhead, while LCAS changes the 
>    SDH/SONET overhead
>    > :-)
>    > >
>    > > LCAS uses a new field in the POH (CTRL) to know when the new VC
>    > transitions
>    > > to a significant payload. LBM proposes to do the same 
>    simply by looking
>    at
>    > > the C2 (for HOVC) transition - regular SDH/SONET field 
>    that will change
>    > > anyway and that is used to indicate the type of payload.
>    > >
>    > > To be more precise about LCAS (I copy from the LCAS 
>    specification,
>    section
>    > > 6.3.1): "The first container frame to contain payload 
>    data for the new
>    > > member shall be the container frame immediately 
>    following the container
>    > > frame that contained the last bit(s) (i.e. the CRC) of 
>    the control
>    packet
>    > > with NORM/EOS message for that member" (the LCAS 
>    control packet).
>    > >
>    > > So, LCAS waits for the last bit of the appropriate 
>    LCAS control packet
>    in
>    > > the POH with the appropriate transition of its CTRL 
>    field. LBM waits
>    > simply
>    > > for the appropriate C2 (for HOVC) or V5 (for LOVC) 
>    transition in the POH
>    > > (don't change SDH/SONET).
>    > >
>    > > By the way, this is version 0 of the draft and it was 
>    just posted.
>    > >
>    > > Suggestions, technical comments and improvements are welcome.
>    > >
>    > > I would appreciate if people could first read and 
>    speak before shooting
>    > > directly :-)
>    > >
>    > > Kind regards,
>    > >
>    > > Eric
>    > >
>    > > -----Original Message-----
>    > > From: Stephen Trowbridge [mailto:[e-mail]]
>    > > Sent: Friday, November 16, 2001 7:49 PM
>    > > To: Mannie, Eric
>    > > Cc: 'Maarten Vissers'; ccamp; q11/15; t1x1.5; Dimitri 
>    Papadimitriou
>    > > Subject: Re: LCAS and draft-mannie-ccamp-gmpls-lbm-tdm-00.txt
>    > >
>    > > Eric,
>    > > The point of my earlier email (items again below, this 
>    time with
>    > > numbers) was not to prompt the comment "#3 is out of 
>    scope, therefore
>    > > it is a choice between #1 and #2" (my parsing of 
>    recent statements).
>    > >
>    > > The issue (and I think the reason for Maarten's strong 
>    reaction) is
>    this:
>    > > GMPLS-LBM and LCAS are NOT different solutions to the 
>    same problem.
>    > > GMPLS-LBM and LCAS ARE solutions to different problems.
>    > > (I hope this is not too subtle for non-native English 
>    speakers).
>    > >
>    > > LCAS is a method to hitlessly change the mappings at the two
>    > > ends of a virtually concatenated trail to increase or 
>    decrease the
>    > > bandwidth. It does not say anything about how components to be
>    > > added or removed from the virtually concatenated "bundle" are
>    > > set up or taken down.
>    > >
>    > > LBM talks about how to set up or take down those added 
>    or removed
>    > > components, but not about the details of changing the 
>    mapping (other
>    > > than that it is done after the new components are successfully
>    > > established.
>    > >
>    > > If you use LBM and change the mappings at the two ends 
>    without some
>    > > coordination such as LCAS, you will have a glitch to 
>    the traffic
>    > > (I think this is simple enough).
>    > >
>    > > The compare/contrast discussion of LBM and LCAS is 
>    appropriate not
>    > > because LCAS is "sacred", but because you are 
>    comparing apples to
>    > > oranges. Statements in this document such as "One 
>    drawback of LCAS is
>    .."
>    > > are unnecessary and inappropriate in this context, 
>    since LBM does
>    > > NOT solve the problem LCAS was designed to solve. 
>    Perhaps if you
>    > > were solving the SAME problem better, this would be 
>    OK, but this
>    > > is not the case.
>    > >
>    > > Time to try to be constructive:
>    > > - I think most of the current discussion of LCAS 
>    should be removed from
>    > > this document. It is irrelevant since LCAS solves a 
>    different problem
>    than
>    > > LBM. To say that one is "better" than the other would 
>    only be meaningful
>    > > if they were solving the same problem (can you really 
>    say that LBM
>    solves
>    > > problem A better than LCAS solves problem B?).
>    > > - LCAS can be a good informative reference as a way to 
>    hitlessly modify
>    > > the size of a virtually concatenated trail.
>    > > - Clarify in the scope that the current draft 
>    describes the use of LBM
>    > > without LCAS, which results in a traffic glitch when 
>    bandwidth is
>    > modified.
>    > > Indicate that the use of LCAS to hitlessly negotiate 
>    the changes to
>    > > the size of virtually concatenated pipes after LBM sets up new
>    > component(s)
>    > > or before LBM takes down old component(s) is a topic 
>    for further study
>    > > and perhaps to be addressed in a future version of the draft.
>    > >
>    > > Regards,
>    > > Steve
>    > >
>    > > Steve Trowbridge wrote:
>    > > > 1. Can you use LBM without LCAS? Yes, if you don't 
>    mind a glitch to
>    > > > the traffic.
>    > > >
>    > > > 2. Can you use LCAS without LBM? Yes, if you don't 
>    care whether the
>    > > > control plane knows that the different LSPs are part 
>    of the same
>    > > > circuit.
>    > > >
>    > > > 3. Can you use LCAS WITH LBM? Hopefully yes, and 
>    this would seem to
>    > > > be the preferred method for many applications.
>    > >
>    > > "Mannie, Eric" wrote:
>    > > >
>    > > > Hello Maarten,
>    > > >
>    > > > >Any sentence in which LCAS and ASON/GMPLS are 
>    compared shows a lack
>    of
>    > > > >understanding of LCAS's objective (my very strong opinion).
>    > > >
>    > > > Sorry, do you mean that it is forbidden to speak 
>    about your LCAS
>    > document
>    > > > and to compare GMPLS and LCAS functionalities ?
>    > > >
>    > > <snip, to save BW>
>