MPLS WG Meeting          IETF72 Dublin           Monday, July 28, 2008



1.  Agenda Bashing

Several drafts were removed from the agenda due to the fact that the 
authors were not able to attend this meeting.  

  The Use of Entropy Labels in MPLS Forwarding 
     <draft-kompella-mpls-entropy-label-01.txt>
 
  Signaled PID When Muxing Payloads over RSVP-TE LSPs 
     <draft-ali-mpls-sig-pid-multiplexing-case-00.txt>

  Signaling RSVP-TE P2MP LSPs in an Inter-domain Environment
     <draft-ali-mpls-inter-domain-p2mp-rsvp-te-lsp-00.txt>

Additionally, the aggregate-fec draft was removed prior to the
meeting.  The most recent draft is just a keep-alive with a few minor
edits.  An updated draft will be presented at the next meeting.


2.  Working group status

http://www1.tools.ietf.org/wg/mpls/  
Loa / George

Loa presented Working Group Status: 

The MPLS WG was re-chartered to allow MPLS-TP work. This will be a
coordinated effort among MPLS, CCAMP, PWE3, and L2VPN. Another
re-chartering effort for other MPLS work will commence after this
meeting.  The target is to complete this by the Minneapolis meeting.

New RFC: 
RFC 5283 "LDP Extension for Inter-are Label LSPs"

RFC Editor Queue: 
- upstream-label
- multicast-encaps

IESG Processing: 
- ldp-igp-synch
- p2mp-te-mib
- ldp-capabilities
- number-0-bw-te-lsps

Active & on agenda:
- remote-lsp-ping
- mpls-and-gmpls-security-framework
- ldp-end-of-lib
- cosfield-def
- soft-preemption
- 3209-patherr

Active:
- rsvp-upstream
- ldp-upstream
- explicit-resource-control-bundle
- fastreroute-mib
- lsp-ping-enhanced-dsmap
- te-scaling-analysis
- rsvp-te-no-php-oob-mapping
- ldp-typed-wildcard
- p2mp-te-bypass
- mp-ldp-reqs
- aggregate-fec

MPLS-TP

There are currently about 10 drafts on MPLS-TP in this meeting.  We
expect another 10 for next meeting.  These will be addressed in a
session on Wednesday joint with PWE3, CCAMP, and L2VPN. The draft
naming convention is that all draftw will contain "mpls-tp" so you can
find them.  An additional WG name indicates which WG is intended to
progress the draft.


3.  Soft preemption, patherr and fast reroute request
    <draft-ietf-mpls-3209-patherr-02.txt>
    <draft-ietf-mpls-soft-preemption-10.txt>
    JP Vasseur

Definitions of soft & hard preemption presented.

Patherr clarifies hard pre-empt and further specifies which codes are
fatal vs non-fatal.

JP proposed that patherr be progressed as a BCP.

For soft-preemption, JP proposed to align the draft with existing
mechanisms as follows. Remove the use of RRO sub-object.  Use Path
Error instead.  Keep the definition of the new SESSION-ATTRIBUTE flag
(or move it to LSP-ATTRIBUTE).  This segued into the next
presentation.


4.  GMPLS LSP Reroute

    Lou Berger
    <draft-berger-mpls-gmpls-lsp-reroute-00.txt>

MPLS and CCAMP are working on similar problem - soft-preemption and
graceful shutdown.  This draft describes how these can be accomplished
with existing protocol mechanisms.  There is no need for additional
mechanisms.

Lou proposed this as a WG doc on the BCP track so that new re-route
proposals can refer back to this description.  Need to align
graceful-shut and soft-preemption.

CCAMP WG chair asked to comment

Adrian F: as far as CCAMP is concerned need to make sure we are
aligned, but need to be quite fast as CCAMP is close to
completion. There appears to be duplication between exiting RFC and
soft-preemption - should get down to one mechanism?

Lou: The duplication exists between the existing RFCs, not this one.

Loa: Should MPLS - bring this draft in as a WG draft? - Not many read
the draft - issue for the list

George: Even though the number is small, those people who are closely
involved appear to be converging on a solution.  Hope to move this
work forward soon to prevent entropy.

Loa: WG status proposal will be discussed on list. Corrections to 
overlapping drafts can start immediately


5.  MPLS and GMPLS Security Framework
  
   <draft-ietf-mpls-mpls-and-gmpls-security-framework-03.txt>
   Luyuan Fang

Changes in 03 draft as a result of review comments
  - Addressed Scott Brim gen-art comments
  - upstream label application - Stephen Farrell and Ross Callon
      comments 
  - added security impact for upstream (a new security mechanism is
      not needed)
  - add IPv6 to filtering in addition to IPv4

Planned Changes: 

  1) GMPLS data plane security is needs to be addressed due to date
     plane / control plane decoupling. Data plane may be accidently
     disconnected without causing a fault to the control plane.

  2) a reference and a brief description on applicability of
     groupkeying for RSVP

Next steps: 
  Update draft as described 
  Request last call in MPLS and CCAMP


6.  LDP support for point to multipoint
    <draft-ietf-mpls-ldp-p2mp-05.txt>
    Ice Wijnands

MP2MP - ordered mode
Need to set up paths in order so that leaf knows the upstream path is
complete.  No additional messages

Micro-loops may form.  However, if there is no IGP inconsistency there
is no micro-loop downstream micro-loop is less harmful if there is a
single injection point upstream - there are multiple injection points

Add Path-Vector TLV - to check for loops - forwarding is not updated
unless PV TLV is loop free.  Only a partial solution - p2pm is
difficult problem

George: how does this impact convergence?

Ice: as soon as the IPG converge happened, or LDP re-converge, then
the loop is gone.

George: need to simulate to understand impact on convergence

Lou: If you are going to simulate, please bring results to the WG

Stewart Bryant: ThereÕs also work in IEEE, check what they are doing.
They propose a hash of the LSP IDs in the routing DB to check
consistency.  It is interesting. Further simulation work needed.


7.  Renaming the Shim Header EXP Field
   
    <draft-ietf-mpls-cosfield-def>
    Loa Andersson

This is now a WG draft.  Renaming the Shim exp field to clarify to
other SDOs that this is NOT experimental!  The field was originally
called CoS, but at time RFC3032 publication CoS was not finalized, so
it was called EXP - but intended for CoS experiments! Rename to CoS to
clarify this.

WGLC will be done in tsvwg, ccamp, pwe3 l2vpn

The draft was used as a test of MPLS-TP review procedure.  Acked by
ITU-T.  Publication will be requested when we resolve name.

A few comments - some on names

CoS, ToS, Traffic mgt, MPLS DS, PHB

But comments all say - don't care much but suggest this

Issue is whether name restricts application - for example ECN/PCN

George: Every term in this space is overloaded - every terms seems to
have some techno-political consequence.  We need to make sure the
document is explicit what is included and not included as valid uses
of the field.

Dave McDyson: The name change is only to allow us to align with RFCs -
but note we are still experimenting

George: We know it's a COS constrained experiment.  Other SDOs do not
know the parameters of the ÒexperimentÓ and that caused problems

Poll of room
Support: 40 to low 50 say CoS is good enough 
Oppose:   1

Will poll on MPLS list

Loa: ask to move forward?
Ross: Yes, proposed standard


8.  LDP End-of-LIB
  
    <draft-ietf-mpls-ldp-end-of-lib-00.txt>
    Rajiv Asati
  
Now a WG doc

There are cases where it is useful for LDP to signal that all labels
sent - for example for graceful restart.

Need a capability message and status notification.

Proposal: 
  - Status TLV with a end-of-lib status code
  - FEC TLV with a typed wildcard FEC element

Changes: must notify end of null list

Now ready for WGLC

Loa: got couple of comments back before WG last call
Stewart: want the draft to reviewed in PWE WG
George: We can last call in both MPLS and PWE WGs


9.  Proxy LSP Ping
   
    <draft-ietf-mpls-remote-lsp-ping-01.txt>
    George Swallow

Motivation - Primarily for P2MP LSPs - scalability and ability to
trace from leaf to source (for mLDP)

Adds 2 msgs - Proxy echo req and proxy echo repl

Ping contains info to get msg back
Allows you to probe closer to problem

A suggestion was made to add label stack.  This could be useful but
also raises some serious security concerns For now, the draft only
supports FECs that are understood by proxy router - need a very good
reason to do otherwise

Changes to this draft

DSCP:
  Can be specified in proxy echo request.  Flag says if this is
  requested Proxy LSR should use this, but may choose to use a
  different value.  Must notify if another value is used

Previous hop object: 
  Local address field added.  Needed to communicate the address by
  which the PHOP knows this node Generally this would be the address
  used in the protocol by which the label for the bottom most FEC was
  exchanged

Next hop object:
  Now 6 address types - ip4 and ip6, plus four formats in DS-Map

Next steps - no functional changes expected, will update with missing
sections, will request WG LC, and would like security review early
WGLC at next IETF


10. MPLS TE: Changing Overbooking factors
    
    <draft-newton-mpls-te-dynamic-overbooking-00.txt>
    Jonathan Newton

In the current design the over-booking factor is fixed - needs to be
altered during failure.  Idea is to change maximum booking under
failure.  For example, change from 60% under normal circumstances to
80% during failure.

Use DSTE to advertise different bandwidth constraints (BCs)

Identify failure mode at Head-End (from for example LSP)
Normal operation use BC1

Change to BC0 on LSP when failure notified to head end

When link comes up - make before break - set up BC1 on LSP comming up, 
then use then reduce other LSP to BC1

Propose BCP - no protocol changes needed

Get f/b from list and then request WG draft status

Francois: some comments that will send to list + plus comment I
missed

Dave McDysan: It would be more comfortable with a info RFC. You could
do local computation 
 
Johnason: I do not want to it do locally; want to allow other nodes to
participate

George: Other thing. Russian Doll model, MAM (GCAC), have you studied?

Johnason: difficult to implement MAM.

Nitin: BC1, BC0, why not have a low priority LSP be preempted 

Johnason: LSPs are all equal priority, cannot use that.

Nitin: Reduce the LSP BW, operator can see the prob. Happening in the
middle of the network.

George: As chair Informational may be more approprate.  Differences
in SP policy are why this needs to be informational rather than BCP

Johnason: agree to work as informational, allow multiple mechanisms to
be used as solutions to the problem.


11. Graceful Shutdown of LDP Adjacency
    
    <draft-boutros-mpls-ldp-gs-adj-00.txt>
    Sami Boutros

2 LSR with two interfaces connected in parallel

2 LDP hello adj, one LDP tcp session

Want to shut down a single link gracefully without affecting traffic
on the other link.

Currently if LSR1 removes one link, lsr2 blackholes traffic on that
link until the LDP hello times-out.

Propose that when LSR1 sends hello msg saying it will terminate adj,
so lsr2 knows not to use it

New TLV - carries R bit - request and ack of graceful shutdown

Future enhancement - add capability advert

Nitin: why can't we just withdraw label?

Sami: Because all labels will be advertised on both links and we still
want MPLS connectivity - we are not stopping MPLS - just removing ECMP

Next step comments from list and request WG draft


12. Requirements for LER Forwarding of IPv4 Option Packets 

    <draft-dasmith-mpls-ip-options-00.txt>
    Dave Smith

Forwarding behavior is IP option dependent Behavior is also
implementation dependent (except for VPNs) Packets that bypass MPLS
are a security issue. Also can trigger imposition of RA Label -
leading to DoS attack on a downstream LSR

Draft specifies that the Ingress LER must (default behavior) make
imposition decision independent of options.  Label Stack values must
be invariant wrt to IP options. How LSRs process options is out of
scope.

On Egress LER - only process IP options when fwd decision based on IP
header.

Request WG draft

Francois: Good problem to tackle. Interested in RSVP implications.
Also need to talk about MPLS VPN to make it explicit

Dave Smith: next version will add a clause that says this will not
apply to RSVP and other control protocols

Loa: How many have read - not many.  Author needs to invite comments
on list and then we will consider as WG draft

End of meeting