[Teas] Comments on draft-ietf-teas-gmpls-lsp-fastreroute

["Matt Hartley (mhartley)" <[e-mail]>]

All,

Some comments on this.

In general: post -> after. It's just easier to read (IMO :)

You don't need to expand MP every time you use it. I lost count of the number of places you've done this.

The document talks about PSC LSPs. Presumably it doesn't apply to LSPs with any other switching capability, but it might be a good idea to explicitly say so (maybe around section 3, where you're setting out other things this document doesn't apply to).

Section 4.1 Does RFC 4090 specify that the upstream label MUST be added to the Path RRO? If not, this draft needs to.

Section 4.2: Similarly, does 4561 specify that the node-id MUST be added? Again, this draft relies on it.

Does this draft work on the assumption that the backup tunnel goes in the same direction as the primary LSP that it protects (i.e. that the backup's head is at the upstream PLR rather than the downstream PLR)? I get the feeling it does (more on this later) but it's not explicitly stated anywhere. If this is the case, it means that perfectly good backup tunnels that happen to be running in the wrong direction can't be used.

4.4.1: This means that a Path message is going to be regenerated every time there's a backup assignment. Given that backup assignments are generally done during Resv processing, this means that if every hop can be a PLR, you're going to get a *lot* of updated Path messages while the LSP is being established. Is this a problem? Can it be mitigated if so? It would be good to see some discussion of this.

4.4.1: this doesn't say that a downstream MP needs to examine the entire Path RRO and look at all BYPASS_ASSIGNMENT objects; it probably should do. It should probably also say that the choice of backup tunnel (if multiple choices exist) is discussed in section 4.4.2.

As written, the draft prevents an upstream backup being assigned unless that backup is also assigned in the downstream direction (4.4.1). The reverse, however, does not apply; when a downstream PLR assigns a backup, there's no guarantee that this backup will also be used for upstream traffic. Is this a problem? And if not, why shouldn't an upstream PLR be permitted to assign an upstream backup in its own initiative if it wants to? Section 4.4.2 seems quietly resigned to the fact that some links may be unprotected in the reverse direction. Is this inevitable?

4.4.3: is the data present in the BYPASS_ASSIGNMENT subobject sufficient to uniquely identify the backup tunnel? I see two issues here.

First, the tunnel-id alone is sufficient only if the source and destination are already known from the RRO, and that's only possible if we assume the source is the node that inserted the BYPASS_ASSIGNMENT subobject and the destination is the processing MP. As mentioned earlier, a backup tunnel originating at the MP could also do the job (these are bidirectional tunnels, after all) but this is precluded at the moment. I think this limitation needs to be justified if you wish to impose it.

Second: even if we only allow backups to originate on the downstream PLR, do we also need the extended tunnel id as an additional discriminator?

Section 5 seems to be based on the assumption that all link failures are bidirectional. Is this reasonable? In particular, what if a link failure is only detected by the upstream PLR?

5.1, at the end: does the upstream PLR have to wait until it receives Path state over the bypass tunnel before redirecting Resv state onto the bypass, or can it redirect Resv state as soon as it detects a failure? It strikes me that the latter might be a good idea, especially if unidirectional failures are a possibility.

6.2: Under what circumstances could the PRR fail to find a reverse bypass tunnel? Given that it became a PRR because it was already a MP, doesn't it always have a suitable bypass? If it didn't, it would be a MP...

6.2: this example focuses on what R3 and R5 are doing. Could you also describe the expected behavior of R2 and R4?

Finally... some nits/readability suggestions. I've pasted significant chunks of the draft below with my suggested changes inline: look for 'MRH'.

Cheers

Matt

Abstract

   This document defines Resource Reservation Protocol - Traffic
   Engineering (RSVP-TE) signaling extensions to support Fast Reroute
   (FRR) of Packet Switched Capable (PSC) Generalized Multi-Protocol
   Label Switching (GMPLS) Label Switched Paths (LSPs).  These signaling
   extensions allow the coordination of bidirectional bypass tunnel

MRH: ...coordination of a bidirectional...

   assignment protecting a common facility in both forward and reverse
   directions of a co-routed bidirectional LSP.  In addition, these
   extensions enable the re-direction of bidirectional traffic and
   signaling onto bypass tunnels that ensure co-routedness of data and
   signaling paths in the forward and reverse directions after FRR to
   avoid RSVP soft-state timeout.

(snip)

1.  Introduction

   Packet Switched Capable (PSC) Traffic Engineering (TE) tunnels are
   signaled using Generalized Multi-Protocol Label Switching (GMPLS)
   signaling procedures specified in [RFC3473] for both unidirectional
   and bidirectional LSPs.  Fast Reroute (FRR) [RFC4090] has been widely
   deployed in the packet TE networks today and is preferred for TE

MRH: preferred -> desirable. 'preferred' implies it exists already, and if that were the case you wouldn't need this draft :)

   GMPLS tunnels.  Using FRR methods also allows to leverage existing

MRH: ...Using FRR procedures also allows the leveraging of... (or maybe ...re-use of...)

   mechanisms for failure detection and restoration in the deployed
   networks.

MRH: ...in deployed networks.

   FRR procedures defined in [RFC4090] describe the behavior of the

MRH: The FRR procedures...

   Point of Local Repair (PLR) to reroute traffic and signaling onto the
   bypass tunnel in the event of a failure for unidirectional LSPs.
   These procedures are applicable to unidirectional protected LSPs
   signaled using either RSVP-TE [RFC3209] or GMPLS procedures
   [RFC3473], however don't address issues that arise when employing FRR

MRH: ...[RFC3473], but they do not address...

   for bidirectional co-routed GMPLS Label Switched Paths (LSPs).

   When bidirectional bypass tunnels are used to locally protect
   bidirectional co-routed GMPLS LSPs, the upstream and downstream PLRs
   may independently assign different bidirectional bypass tunnels in
   the forward and reverse directions.  There is no mechanism in FRR

MRH: ...in the FRR procedures...

   procedures defined in [RFC4090] to coordinate the bidirectional
   bypass tunnel selection between the downstream and upstream PLRs.

   When using FRR procedures with bidirectional co-routed GMPLS LSPs, it
   is possible in some cases (e.g. when using node protection bypass
   tunnels post a link failure event and when RSVP signaling is sent
   in-fiber and in-band with data), the RSVP signaling refreshes may
   stop reaching some nodes along the primary bidirectional LSP path
   after the PLRs complete rerouting traffic and signaling onto the
   bypass tunnels.

MRH: maybe replace this lot with:

When using FRR procedures with bidirectional co-routed GMPLS LSPs, it is possible in some cases for the RSVP signaling refreshes to stop reaching some nodes along the primary LSP path after the PLRs finish rerouting signaling onto the bypass tunnels. This may occur when using node protection bypass tunnels after a link failure event and when RSVP signaling is sent in-fiber and in-band with data.

   This is caused by the asymmetry of paths that may be
   taken by the bidirectional LSP's signaling in the forward and reverse
   directions after FRR reroute.  In such cases, the RSVP soft-state
   timeout eventually causes the protected bidirectional LSP to be

MRH: ...timeout causes...

   destroyed, and consequently impacts protected traffic flow after FRR.

MRH: ...destroyed, with subsequent traffic loss after FRR.

   Protection State Coordination Protocol [RFC6378] is applicable to FRR
   [RFC4090] for local protection of bidirectional co-routed LSPs in
   order to minimize traffic disruptions in both directions.  However,
   this does not address the above mentioned problem of RSVP soft-state
   timeout in control plane.

   This document proposes solutions to the above mentioned problems by
   providing mechanisms in the control plane to complement FRR

MRH: ...complement the FRR procedures...

   procedures of [RFC4090] in order to maintain the RSVP soft-state for
   bidirectional co-routed protected GMPLS LSPs and achieve symmetry in
   the paths followed by the traffic and signaling in the forward and
   reverse directions post FRR.  The document further extends RSVP
   signaling so that the bidirectional bypass tunnel selected by the
   upstream PLR matches the one selected by the downstream PLR node for
   a bidirectional co-routed LSP.

   Unless otherwise specified in this document, fast reroute procedures

MRH: ...document, the FRR procedures...

   defined in [RFC4090] are not modified for GMPLS signaled tunnels.

(snip)

3.  Fast Reroute For Unidirectional GMPLS LSPs

   FRR procedures defined in [RFC4090] are applicable to unidirectional

MRH: The FRR procedures...

   protected LSPs signaled using either RSVP-TE or GMPLS procedures and
   are not modified by the extensions defined in this document.  These
   FRR procedures also apply to bidirectional associated GMPLS LSPs
   where two unidirectional GMPLS LSPs are bound together by using
   association signaling [RFC7551].

4.  Bypass Tunnel Assignment for Bidirectional GMPLS LSPs

   This section describes signaling procedures for bidirectional bypass
   tunnel assignment for GMPLS signaled PSC bidirectional co-routed TE
   LSPs.

4.1.  Merge Point Labels

   To correctly reroute data traffic over a node protection bypass
   tunnel, the downstream and upstream PLRs have to know, in advance,
   the downstream and upstream Merge Point (MP) labels so that data in
   the forward and reverse directions can be tunneled through the bypass
   tunnel post FRR respectively.

MRH: .... can be redirected through the bypass tunnel after FRR.

   [RFC4090] defines procedures for the downstream PLR to obtain the
   protected LSP's downstream MP label from recorded labels in the RRO
   of the RSVP Resv message received at the downstream PLR.

   To obtain the upstream MP label, existing methods [RFC4090] to record

MRH: ...label, the procedures defined in [RFC4090] to record the upstream...

   upstream MP label are used in the RRO of the RSVP Path message.  The
   upstream PLR can obtain the upstream MP label from the recorded label
   in the RRO of the received RSVP Path message.

4.2.  Merge Point Addresses

   To correctly assign a bidirectional bypass tunnel, the downstream and
   upstream PLRs have to know, in advance, the downstream and upstream
   Merge Point (MP) addresses.  [RFC4561] defines procedures for the PLR
   to obtain the protected LSP's merge point address in multi-domain
   routing networks where a domain is defined as an Interior Gateway
   Protocol (IGP) area or an Autonomous System (AS).

   [RFC4561] defines procedures for the downstream PLR to obtain the
   protected LSP's downstream merge point address from the recorded
   node-IDs in the RRO of the RSVP Resv message received at the
   downstream PLR.

MRH: these two paragraphs repeat a lot of stuff. Merge them?

   To obtain the upstream MP address, existing methods [RFC4561] to

MRH: ...address, the procedures specified in [RFC4561] to...

   record upstream MP node-ID are used in the RRO of the RSVP Path
   message.  The upstream PLR can obtain the upstream MP address from
   the recorded node-IDs in the RRO of the received RSVP Path message.

4.3.  RRO IPv4/IPv6 Subobject Flags

   RRO IPv4/IPv6 subobject flags are defined in [RFC4090], Section 4.4
   and are applicable to the FRR procedure for the bidirectional GMPLS

MRH: ...for bidirectional...

   tunnels.

   [RFC4090] defined procedure is used by the downstream PLR

MRH: The procedure defined in [RFC4090] is used...

   independently to signal the Ipv4/IPv6 subobject flags in the RRO of

MRH: Ipv4 -> IPv4

   the RSVP Path message.  Similarly, this procedure is used by the
   upstream PLR independently to signal the IPv4/IPv6 subobject flags in
   the RRO of the RSVP Resv message.

4.4.  Bypass Tunnel Assignment Co-ordination

   This document defines signaling procedure and a new BYPASS_ASSIGNMENT

MRH: procedure -> procedures

   subobject in RSVP RECORD_ROUTE object used to co-ordinate the

MRH: ...in the RSVP RECORD_ROUTE object...

   bidirectional bypass tunnel selection between the downstream and
   upstream PLRs.

4.4.1.  Bypass Tunnel Assignment Signaling Procedure

(snip)

   The upstream PLR (downstream MP) that detects a BYPASS_ASSIGNMENT
   subobject, whose bypass tunnel and the node-ID subobject when used as
   a "bypass tunnel source" terminates locally, assigns the matching
   bidirectional bypass tunnel in the reverse direction, and forwards
   the RSVP Path message downstream.  Otherwise, the bypass tunnel
   assignment subobject is simply forwarded downstream along in the RSVP
   Path message.

MRH: This paragraph is a bit garbled. Can you rewrite to make it clearer?

   Bypass assignment co-ordination procedure described above can be used

MRH: The bypass assignment...

   for both one-to-one backup described in Section 3.1 of [RFC4090] and
   facility backup described in Section 3.2 of [RFC4090].

4.4.2.  Bypass Tunnel Assignment Policy

   In the case of upstream PLR receiving multiple BYPASS_ASSIGNMENT
   subobjects from multiple downstream PLRs, the decision of selecting a
   bypass tunnel in the reverse direction can be based on a local
   policy, for example, prefer link protection versus node protection
   bypass tunnel, or prefer the most upstream versus least upstream node
   protection bypass tunnel.

MRH: "the selection of a bypass tunnel in the reverse direction can be based on local policy. Examples of such a policy could be to prefer link protection over node protection, or to prefer the bypass tunnel to the furthest upstream node."

                              However, it is recommended that nodes

MRH: recommended -> RECOMMENDED? Or if you don't mean to use 2119 language here, could you reword this? It's ambiguous as it stands.

   along the LSP path employ identical policy for bypass tunnel
   assignment.


   When different policies are used for bypass tunnel assignment on the
   LSP path, it may result in some links in the reverse direction not
   assigned bypass protection as shown in examples below.

   As shown in Example 1, node A assigns a node protection bypass tunnel
   in the forward direction but node C does not assign a node protection
   bypass tunnel in the reverse direction for a protected bidirectional
   GMPLS LSP.  Both nodes B and C assign a link protection bypass
   tunnel.  As a result, there is no fast reroute protection available
   in the reverse direction for link A-B for this LSP.


                      +------->>------+
                     /          +->>-+ \
                    /          /      \ \
                   /          /        \ \
                  A -------- B --------- C
                              \        /
                               \      /
                                +-<<-+

         Example 1: An example of different bypass assignment policy

MRH: add an arrow to indicate the direction of the ABC LSP. Same in example 2 below.

(snip)

4.4.3.  BYPASS_ASSIGNMENT Subobject

   The BYPASS_ASSIGNMENT subobject is used to inform the MP of the

MRH: ...inform the downstream MP...

   bypass tunnel being assigned by the PLR.  This can be used to
   coordinate the bypass tunnel assignment for the protected LSP by the
   downstream and upstream PLRs in the forward and reverse directions
   respectively prior or post the failure occurrence.  This subobject
   SHOULD only be inserted into the RSVP Path message by the downstream
   PLR and MUST NOT be changed by downstream LSRs.

MRH: maybe... "This subobject SHOULD be inserted into the Path RRO by the downstream PLR. It SHOULD NOT be inserted into an RRO by a node which is not a downstream PLR. It MUST NOT be changed by downstream LSRs and MUST NOT be added to a Resv RRO."


   The BYPASS_ASSIGNMENT subobject in RRO has the following format:

          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    |      Length   |      Bypass Tunnel ID         |
       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+


      Type

            Downstream Bypass Assignment.

MRH: maybe mention that the value is TBD pending IANA assignment?

(snip)

5.  Link Protection Bypass Tunnels for Bidirectional GMPLS LSPs

   When a bidirectional link protection bypass tunnel is used, after a
   link failure, downstream PLR reroutes RSVP Path and traffic over

MRH: failure... the downstream PLR reroutes traffic and RSVP messages over the bypass tunnel using the procedures defined in...

   bypass tunnel using procedures defined in [RFC4090].  Upstream PLR
   may reroute traffic and RSVP Resv upon detecting the link failure or
   upon receiving RSVP Path message over a bidirectional bypass tunnel.

MRH: Upstream PLR behavior is unclear here. I assume you mean that it MAY use whichever trigger (Path message or link-down) it gets first? You should probably say that it MUST do one or the other.

   This allows both traffic and RSVP signaling to flow on symmetric
   paths in the forward and reverse directions of a bidirectional
   tunnel.

(snip)

   Consider the Traffic Engineered (TE) network shown in Figure 1.
   Assume every link in the network is protected with a link protection
   bypass tunnel (e.g. bypass tunnel T3).  For the protected
   bidirectional co-routed LSP whose (active) head-end is on router R1
   and (passive) tail-end is on router R5, each traversed router (a
   potential PLR) assigns a link protection bidirectional co-routed
   bypass tunnel.

MRH: are active and passive industry-standard terms? I'd have thought using head and tail would suffice.

(snip)

6.  Node Protection Bypass Tunnels for Bidirectional GMPLS LSPs

(snip)

   Consider the Traffic Engineered (TE) network shown in Figure 2.
   Assume every link in the network is protected with a node protection
   bypass tunnel.  For the protected bidirectional co-routed LSP whose
   (active) head-end is on router R1 and (passive) tail-end is on router
   R6, each traversed router (a potential PLR) assigns a node protection
   bidirectional co-routed bypass tunnel.

MRH: Again, are active and passive industry-standard terms?

(snip)

6.2.  Behavior Post Link Failure To Re-coroute

   The downstream Merge Point (MP) R5 that receives rerouted protected
   LSP RSVP Path message through the bypass tunnel, in addition to the
   regular MP processing defined in [RFC4090], gets promoted to a Point
   of Remote Repair (PRR role) and performs the following actions to
   re-coroute signaling and data traffic over the same path in both
   directions:

      o Finds the bypass tunnel in the reverse direction
        that terminates on the Downstream PLR R3.  Note: the Downstream
        PLR R3's address is extracted from the "IPV4 tunnel sender
        address" in the SENDER_TEMPLATE object.

MRH: bypass tunnel's SENDER_TEMPLATE

(snip)

   If downstream MP R5 receives multiple RSVP Path messages through
   multiple bypass tunnels (e.g. as a result of multiple failures), the
   PRR SHOULD identify a bypass tunnel that terminates on the farthest
   downstream PLR along the protected LSP path (closest to the primary
   bidirectional tunnel head-end) and activate the reroute procedures
   mentioned above.

MRH: Don't you mean the furthest *upstream* PLR?

(snip to end)