Response to your questions about GMPLS parameters

["Adrian Farrel" <adrian@olddog.co.uk>]

To: Jim Jones, OIF Technical Committee Chair
From: Adrian Farrel and Kireeti Kompella,
          WG Co-Chairs for IETF CCAMP
Copy: Alex Zinin and Bill Fenner, IETF Routing Area Directors
Subject: Response to your questions about GMPLS parameters.

Dear Jim,

Thanks for your correspondence about the questions with respect to GMPLS
parameters that arose before and during your interoperability testing.
CCAMP is pleased to receive such questions and is glad to have the
opportunity to explain the intended operation of the GMPLS protocols.

Much of the material supplied below can be simply extracted from the
relevant RFCs.

> 1. Use of the NCC and RCC fields for STS-3c/VC-4 connections
>
> During OIF testing it was noted that some ambiguity exists in the
> specification of encoding of NCC, RCC and NVC for certain types of
> connections: NCC and RCC for an STS-3c/VC-4 connection can be set to 0
> or to 1 depending on which example of RFC 3946 is followed.
>
> Clarification is requested from IETF CCAMP as to which setting is
> considered correct, or if both settings should be accepted (this
> procedure was used during testing at Supercomm).

This question about RFC 3946 was raised informally on the CCAMP mailing
list at the start of March this year.

The intention of the editors of RFC 3946 was that a VC-4 elementary signal
would be represented using RCC=0 and NCC=0, while an STS-3c SPE elementary
signal would have RCC=1 and NCC=1. This follows the examples in the annex.

Your question probably arises from the two notes and subsequent paragraph
in section 2.1 or RFC 3946. The text here is compatible with the annex,
but culminates with...

"A RCC value different from 0 must imply a number of contiguous components
greater than 1."

The editors intended this to mean "greater than or equal to 1."

As a result of your question there has been some heated debate within
CCAMP driven by the fact that some people have implemented according to
the word of the RFC while others have implemented according to the
intention of the editors (using the examples in the annex to drive their
work).

We plan to issue a revision to RFC 3946 which will include the following
compromise position.

a. "greater than" will be changed to "greater than or equal to"
b. A third note will be added to section 2.1 to clarify the expected
behavior as follows...

 Note 3: Following these rules, when requesting a VC-4 signal, the
 RCC and the NCC values are set to 0, whereas for an STS-3c SPE
 signal, the RCC and the NCC values are set to 1. However, since
 the signals are indistinguishable in the data plane, the requesting
 upstream node MAY set the RCC and NCC values to either both
 0 or both 1 without impacting the function. The downstream node
 MUST accept both possible settings of RCC and NCC and handle
 them interchangeably. The upstream node SHOULD be consistent
 in its choice of RCC and NCC value settings for all LSPs requested
 on any one interface.

This solution is still subject to agreement from the CCAMP community, but
appears to reflect a position which facilitates maximum interworking.

> 2. Setting of NVC for VCAT connections
>
> It was also noted that the setting of NVC may be somewhat ambiguous for
> the case where diverse connections are used within a single VCAT group.
> Each individual RSVP session controls a single connection, but the
> connection is part of a larger VCAT group and carries VCAT encoding of
> the H4 byte. Clarification is requested from IETF CCAMP and ITU-T
Q.14/15
> as to the correct setting of NVC for this case (0 or 1?). It should be
> noted that this case may occur with a VCAT group with only a single
initial
> member, and that the NVC may provide an indication that VCAT encoding of
> the H4 byte is in use for the connection.

A VCn-Xv group split into X components requires each of its component to
be signaled with the NVC value set to 1. This setting is regardless of how
the components are established.

> 3. Length of the Interface Switching Capability TLV
>
> Although the Interface Switching Capability TLV defined by CCAMP for
> SONET/SDH connections was not used for the testing, it was noted that
> the text describing the length of the Interface Switching Capability TLV
> defined in draft-ietf-ccamp-ospf-gmpls-extensions-12.txt may be slightly
> ambiguous due to the use of padding bytes.
>
> RFC 3630 states that "The TLV is padded to four-octet alignment; padding
> is not included in the length field (so a three octet value would have a
> length of three, but the total size of the TLV would be eight octets)."

Yes. Section 2.3.2 of RFC3630 gives a definitive statement of the meaning
of the length field and the use of padding, and provides an example.

> Reading of the encoding in draft-ietf-ccamp-ospf-gmpls-extensions-12.txt
> specifies that the length of the TLV for TDM is 41 bytes plus 3 bytes of
> padding, and should be given in the length field as 41 bytes rather than
> 44. OIF requests verification of this interpretation from the experts in
> IETF CCAMP group.

Note that the Interface Switching Capability Descriptor defined in
draft-ietf-ccamp-ospf-gmpls-extensions-12.txt is a sub-TLV of the Link
TLV. Sub-TLVs and TLVs follow the same encoding rules.

The ISCD TLV for TDM contains the following fields...
  type       2 bytes
  length     2 bytes
  ---
  switch cap 1 byte
  encoding   1 byte
  reserve    2 bytes
  LSP b/w 0  4 bytes
  LSP b/w 1  4 bytes
  LSP b/w 2  4 bytes
  LSP b/w 3  4 bytes
  LSP b/w 4  4 bytes
  LSP b/w 5  4 bytes
  LSP b/w 6  4 bytes
  LSP b/w 7  4 bytes
  min b/w    4 bytes
  indication 1 byte
            ==
            41 bytes

We presume that your question relates to whether the 3-byte field shown as
"padding" in the TDM-specific figure on page 6 of
draft-ietf-ccamp-ospf-gmpls-extensions-12.txt is an implicit or an
explicit field.

It is an implicit field, and should not be included in the length of the
TLV.

Nevertheless, we take this opportunity to remind the OIF that
implementations of GMPLS protocols should be conservative in what they
send and liberal in what they receive. Thus, an implementation that
receives a TDM ISCD TLV with length 44 should not reject the TLV for this
reason. It should parse the TLV according to the defined fields and skip
the final three bytes. Thus, it should not affect a receiving
implementation if the sending implementation has treated the "padding"
field as implicit or explicit. In the event that a receiving
implementation rejected such a TLV on grounds of the value contained in
the length field being too large, the fault would lie with the receiving
implementation not the sending implementation.

> 4. Use of ADMIN_STATUS in an initial PATH message
>
> Some implementations sent an ADMIN_STATUS object with no flags set in
> the initial PATH message, i.e., when no status change was being
requested.
> Although this did not serve any particular function, it was believed
> that this could be accepted as RFC3473, sect. 7.2 (page 18) states:
>
> "The absence of the object is equivalent to receiving an object
> containing values all set to zero (0)."
>
> It was our interpretation based on this text that a node should accept
> an ADMIN_STATUS object with no flags set in the same way as if
> the object was missing. Comment on this interpretation is welcome.

The effect of the meaning is as you state, but the intention of the
meaning is reversed. That is, an implementation should accept the absence
of the ADMIN_STATUS object in the same way as if the object was present
with no flags set. That is, the default behavior is to consider the
ADMIN_STATUS object as a standard part of the processing.

We note from your first paragraph that you assume that the ADMIN_STATUS
object is used to change the status of the LSP. This is a
misinterpretation - it is used to control the status of the LSP. Thus, if
there is no change to the status of an LSP, refresh messages must continue
to carry the ADMIN_STATUS object with the same bit setting.

In this way, it is not possible to "drop" the ADMIN_STATUS object without
having the same meaning as transmitting the object with all bits cleared.

> 5. Handling of multiple received ResvConf Request objects
>
> When a connection desires a confirmation that the service (i.e.
> connection) requested is in place, a RESV_CONF_REQ object is included in
> the RESV message. As this object is received by the remote end of the
> reservation, it will send a RESV_CONF message back to the requester.
>
> However, it is unclear whether it is necessary to send a RESV_CONF
> message when the RSVP connection state is refreshed by subsequent
> RESV. This becomes potentially burdensome, especially when the
> reservation is being rapidly refreshed. Therefore we ask: should the
> remote end send a RESV_CONF message for subsequent RESV
> messages that still include the RESV_CONF_REQ object? Or is it
> required that the requestor of the reservation remove the
> RESV_CONF_REQ object to prevent the generation of further
> RESV_CONF messages? Comment on this issue from IETF CCAMP
> is requested.

It is fundamental to the implementation of RSVP-TE that there is a good
understanding of the distinction between a trigger message and a refresh
message. This can be achieved by reading section 1.1 of RFC2961.

Following this understanding, you will note that a refresh message does
not cause any processing to be performed at the LSR that receives it (in
this case the ingress). You will also note that refresh processing is not
end-to-end as implied in your text, but is hop-by-hop.

Thus, a downstream LSR that wishes to trigger a new ResvConf message must
make a specific change to the content of the Resv message that it sends in
order to cause a trigger message to be propagated through the network to
the ingress LSR. Such processing is implementation specific but might
include the toggling of the presence of the RESV_CONFIRM object on the
Resv message.

Note that a ResvConf message is not necessarily reliably delivered
end-to-end. Relying on the receipt of a ResvConf message before doing
something (e.g. turning on the laser) might be a poor idea. GMPLS uses the
Administrative Status object and in particular the R-bit in order to
reliably achieve this function.

> 6. Symmetry of Refresh Reduction usage
>
> During interop testing, we ran into a conflict caused by varying
> interpretations of RFC2961, regarding the use of SRefresh messages and
> the Refresh Reduction capabilities of the two ends of a given link. One
> interpretation of RFC2961 indicates that setting the Refresh Reduction
> Capability flag in the RSVP header indicates that that interface shall
> be capable of receiving messages related to Refresh Reduction -
> including the SRefresh message. This would be true even if the other
> end of the link for that interface were NOT indicating Refresh Reduction
> Capability, since the RFC makes no statement about symmetry in this
> matter.
>
> Another interpretation is that both ends of an interface must indicate
> Refresh Reduction Capability before either end can use such messages,
> i.e, use of Refresh Reduction on a link is symmetric.
>
> Comment from CCAMP WG on the correct interpretation is requested.

We are confused by your question.
You correctly state that the use of the refresh-reduction-capable bit
indicates the ability of an LSR to support the receipt of refresh
reduction options and messages. To quote from section 2 of RFC2961...
           When set, indicates that this node is willing and capable of
           receiving all the messages and objects described in this
           document.  This includes the Bundle message described in
           Section 3, the MESSAGE_ID objects and Ack messages described
           in Section 4, and the MESSAGE_ID LIST objects and Srefresh
           message described in Section 5.  This bit is meaningful only
           between RSVP neighbors.
This makes no statement about whether the LSR intends to use these options
when communicating with another LSR.

However, you will note that some refresh reduction procedures require that
a message is sent and response returned. In order to make use of the
response, the receiver must be capable of receiving and processing the
response. Thus, it would be usual for an LSR that is capable of sending
refresh reduction options and messages to also set the
refresh-reduction-capable bit.

In summary:
- An LSR must not send refresh reduction options or messages
  to an LSR that is not setting the refresh-reduction-capable
  bit.
- An LSR may send refresh reduction options or messages
  to an LSR that is setting the refresh-reduction-capable bit.
- An LSR that wishes to successfully use responded refresh
  reduction options or messages should set the refresh-
  reduction-capable bit.

Note, finally, that section 2 of RFC 2961 states that "When it is not
known if a next hop supports the extension, standard Path and Resv message
based refreshes MUST be used."

> 7. Sending of ACKs bundled with the RSVP HELLO
>
> During interop testing, it was observed that Message Acks were
> piggybacked onto RSVP Hello messages, when the receiving end
> was not using the Hello protocol. In this situation, the incoming
> Hello's were discarded and the Acks were lost.
>
> We believe that Message Acks should only be piggybacked onto mandatory
> messages, and not on Hello messages because of this problem. Comment on
> this interpretation is requested.

You use of the terms "bundled" and "piggybacked" are contradictory.

"Bundled" implies the use of the Bundle message.
RFC 2961 states...
   A sub-message MAY be any message type except for another
   Bundle message.
Thus, Ack messages may be bundled with other messages. (Although one might
consider this perverse since the Ack message is only introduced to handle
the case when the Ac/Nack objects have no other message on which they can
be carried.)

Further, RFC 3209 states...
   A Hello message may be included
   as a sub-message within a bundle message.

Therefore, it acceptable for a Ack and Hello messages to be bundled
together.
The processing rules (RFC 29610 for Bundled messages are such that each
sub-message is processed in its own right, and the non-support/non-use of
Hello messages should not impact the processing of other messages.

On the other hand, "piggybacked" implies the use of the Ack/Nack objects
within a Hello message.

Section 4.1 of RFC2961 states that Ack/Nack objects may be included in the
"standard" RSVP messages, and shows where they are placed. However, RFC
3209 defines the Hello message as not including the Ack/Nack objects...

   <Hello Message> ::= <Common Header> [ <INTEGRITY> ]
                              <HELLO>

Since RFC 3209 post-dates RFC 2961, this definition is definitive and the
Ack/Nack objects should not be present on the Hello message.

Give that section 5.3 of RFC 3209 states...
   The Hello Message is completely OPTIONAL.  All messages may be
   ignored by nodes which do not wish to participate in Hello message
   processing.
...it is not particularly important what the message format rules are. An
implementation that chooses to place an Ack/Nack object in a Hello message
knows that the object might be discarded unprocessed.

> 8. TSPEC format to be used for Ethernet connections

The CCAMP working group has discussed the use of GMPLS for control of
Ethernet devices, and this has led to the publication of a new
Internet-Draft entitled "Use of the Generalized Multi-Protocol Label
Switching control plane for point-to-point Ethernet Label Switching" which
you can find at
http://www.ietf.org/internet-drafts/draft-andersson-gels-bof-prep-00.txt.
The Abstract of this draft reads as follows:

   This document proposes starting a work within the IETF to apply the
   Generalized Multi-Protocol Label Switching (GMPLS) control plane to
   Ethernet Label Switching and to make extensions to the GMPLS control
   plane protocols as necessary for this application.  This will be done
   based on the protocols developed by the MPLS and CCAMP working groups
   in the IETF.  Ethernet Label Switching will use the data plane
   encodings as specified by the IEEE 802 standards.

   This document intends to gather the information necessary to have a
   "GMPLS Ethernet Label Switching" BoF in Vancouver.

We are certain that the IETF will want to ensure that the OIF's
requirements for Ethernet are adequately covered by GMPLS and would urge
the OIF to document its requirements and to send them to the IETF for
inclusion. In this instance you may send such requirements to us and we
will ensure that they are passed on to the correct group that undertakes
this work.

Best regards,
Adrian Farrel
Kireeti Kompella