Re: Moving right along ... generalized-signaling-06

[Maarten Vissers <[e-mail]>]

All,

Zhi-Wei Lin wrote:
> 
> Hi,
> 
> I think active/standby and extra-traffic/normal-traffic are also two
> separate descriptors. 

They are indeed separate descriptors. See e.g. draft G.gps.

> Extra traffic may be carried on a standby connection (but a standby 
> connection does not necessarily have extra traffic on it).

Correct. 1+1 prot architecture can't even support extra traffic.
E.g. 1:1 and 1:n architectures can support extra traffic. As a 1:1 architecture
is more complex than a 1+1 scheme, such 1:1 architecture will typically be
deployed when extra traffic support is required. See for
advantages/disadvantages associated with each protection architecture section
6.2 of draft G.gps:

Possible advantages of the 1+1 architecture include:
1) low complexity.
2) for the case of unidirectional switching, the possibility to support dual
node interconnect of protected subnetworks.

Possible disadvantages of the 1+1 architecture include:
3) 100% extra capacity.

Possible advantages of the 1:n, m:n, (1:1)n architecture include:
1) possibility to provide protection access; the protection transport
entity/bandwidth can transport an extra traffic signal during periods the
protection transport entity/bandwidth is not required to transport a normal
traffic signal.
2) extra capacity restricted to 100/n % or m x 100/n %
3) for case of m:n, protection is possible for up to m faults.

Possible disadvantages of the 1:n, m:n, (1:1)n architecture include:
4) complexity.
5) for case of SNC protection class, the need for additional sublayer
termination functions at ingress and egress points of the protected domain on
each working and protection transport entity.
6) does not support dual node interconnect of protected subnetworks.
7) n >= 2: each of the n working transport entities must be routed via different
facilities and equipment to prevent the existence of common points of failure
that can not be protected by the single protection transport entity in a 1:n and
(1:1)n architecture. 

NOTE:	Typically, n+1 alternative paths between two nodes in the network will not
be available. As such, 1:n and (1:1)n, with n >= 2, architectures will not
provide adequate protection for the n normal signals transported normally via
the n working transport entities. n = 1 seems the only reasonable choice.

NOTE:	In ATM, protection access is not explicitly required to allow usage of the
normally unused protection bandwidth; ABR and UBR traffic could use this
protection bandwidth by means of a over-subscription of the bandwidth of the
server signal containing the protection transport entity. The ABR/UBR higher
layer control mechanism is assumed to reduce the traffic when the protection is
actually used. The ingress/egress nodes of the protection domain do not have to
align with ingress/egress nodes of ABR/UBR traffic. This adds flexibility to the
network, and reduces complexity.


> 
> How do we define normal traffic? Is this user traffic carried by working
> path (when working is active), or also carried by recovery path (when
> recovery is active)?

Normal traffic signal is defined as follows (see G.gps):

3.10.1	Normal traffic signal - Traffic signal that is protected by two
alternative transport entities, called working and protection entities.

3.10.2	Extra traffic signal - Traffic signal that is carried over the protection
transport entity and/or bandwidth when that transport entity/bandwidth is not
being used for the protection of a normal traffic signal; i.e. when protection
transport entity is standby. Whenever the protection transport entity/bandwidth
is required to protect or restore the normal traffic on the working entity, the
extra traffic is pre-empted. Extra traffic is not protected.

3.10.3	Null signal - The null signal is indicated on the protection transport
entity if it is not used to carry normal or extra traffic. The null signal can
be any kind of signal that conforms to the signal structure of the specific
layer and is ignored (not selected) at the tail end of the protection.

> 
> And once you defined these, I think you still may need to specify
> whether a connection is revertive or non-revertive.

Revertive and non-revertive are protection "operation types".

> 
> For example, take the case of 1+1. One operator may decide to use
> non-revertive (because both paths are "equivalent") so avoid a switching
> hit on the traffic (order of milliseconds?), while another operator may
> decide to use revertive (because the recovery path is much longer/more
> hops). 

Section 6.4/G.gps lists 5 cases where revertive operation may be appropriate:

1) Where parts of the protection transport entity may be taken to provide
capacity to meet a more urgent need.  For example, where protection transport
entity can be taken out of service to release capacity for use in restoring
other traffic.
2) Where the protection transport entity may be subject to frequent
re-arrangement.  For example, where a network has limited capacity and
protection routes are frequently re-arranged to maximize network efficiency when
changes occur in the network.
3) Where the protection transport entity is of significantly lower performance
than the working transport entity.  For example, where the protection transport
entity has a worse error performance or longer delay than the working transport
entity.
4) When an operator needs to know which transport entities are carrying normal
traffic in order to simplify the management of the network.
5) Where dual node interconnect mechanism is applied to reliably interconnect
two protected subnetworks. 


> Could these decisions be made per protected LSP basis? 

Yes.

> If yes, then would tail-end need to know which mode it is operating/to
> use (the intermediate nodes don't need to know this)?

For the case of 1+1 architecture with "uni-directional switching" (switching
type is another parameter) only the tail end needs to know the operation type.
For a bi-directional connection, each tail end needs to know its operation
type... in principle one direction can use non-revertive whereas the other
direction can use revertive... but this is not logical, so default is to have
both tail ends operating with the same operation type.

For 1+1/1:n "bi-directional switching" both head end and tail ends need to know
operation type, and the operation type must be the same at both ends.
Intermediate nodes don't need to know about any of those protection parameters.
Those protection parameters are allocated to the protection processes at the
edge of the protected domain. See also figure 3 in section 3.2 in VBI
contribution D.301 (SG15, Oct. 2001).

Regards,

Maarten

> 
> Zhi
> 
> Ben Mack-Crane wrote:
> 
> > Maarten,
> >
> > In the work on a recovery framework for mpls (draft-ietf-mpls-recovery-frmwrk-03.txt)
> > we chose "working path" and "recovery path."  We used "recovery" instead of "protection"
> > because it seemed a more neutral term given that we were including both protection
> > switching and reroute recovery mechanisms.
> >
> > We did not address the state (active/standby).  We use the terms "extra traffic" and
> > "working traffic" (perhaps the latter could be changed to "normal traffic").
> >
> > It would be useful in moving forward in CCAMP if we could try to adopt terms
> > that are consistent with those in current use elsewhere.
> >
> > Regards,
> > Ben
> >
> > Maarten Vissers wrote:
> >
> >>There exist well defined protection terminology in ITU-T for the transport
> >>plane. "Working" and "Protection" are being used and not primary/secondary. E.g.
> >>a 1+1 architecture has one working connection, one protection connection and a
> >>permanent bridge.
> >>
> >>Besides working/protection indication for the transport entity, there is
> >>- "active" and "standby" to indicate if the signal is selected from the working
> >>or protection transport entity; i.e. if selector selects from working, the
> >>working is active and protection is standby, if the selector selects from
> >>protection the working is standby and the protection is active.
> >>- "normal" and "extra traffic" signal. The normal signal is protected.
> >>
> >>Regards,
> >>
> >>Maarten
> >>
> >>[e-mail] wrote:
> >>
> >>>Jonathan....review the text below....I think the problem is 1:1.
> >>>
> >>>neil
> >>>
> >>>
> >>>>-----Original Message-----
> >>>>From: Jonathan Lang [mailto:[e-mail]]
> >>>>Sent: 12 December 2001 17:46
> >>>>To: 'Ben Mack-Crane'; John Drake
> >>>>Cc: Lou Berger; Kireeti Kompella; [e-mail]
> >>>>Subject: RE: Moving right along ... generalized-signaling-06
> >>>>
> >>>>
> >>>>Ben,
> >>>>  Please see inline.
> >>>>
> >>>>Thanks,
> >>>>Jonathan
> >>>>
> >>>>
> >>>>>-----Original Message-----
> >>>>>From: Ben Mack-Crane [mailto:[e-mail]]
> >>>>>Sent: Wednesday, December 12, 2001 8:56 AM
> >>>>>To: John Drake
> >>>>>Cc: Lou Berger; Kireeti Kompella; [e-mail]
> >>>>>Subject: Re: Moving right along ... generalized-signaling-06
> >>>>>
> >>>>>
> >>>>>See comment below.
> >>>>>
> >>>>>Regards,
> >>>>>Ben
> >>>>>
> >>>>>John Drake wrote:
> >>>>>
> >>>>>>Snipped
> >>>>>>
> >>>>>>-----Original Message-----
> >>>>>>From: Ben Mack-Crane [mailto:[e-mail]]
> >>>>>>Sent: Tuesday, December 11, 2001 6:38 AM
> >>>>>>To: Lou Berger
> >>>>>>Cc: Kireeti Kompella; [e-mail]
> >>>>>>Subject: Re: Moving right along ... generalized-signaling-06
> >>>>>>
> >>>>>>
> >>>>>>>>7) In Protection Information it states "The resources
> >>>>>>>>
> >>>>>allocated for a
> >>>>>
> >>>>>>>>   secondary LSP MAY be used by other LSPs until the
> >>>>>>>>
> >>>>>primary LSP fails
> >>>>>
> >>>>>>>>   over to the secondary LSP."  This may not always be
> >>>>>>>>
> >>>>>the case.  An
> >>>>>
> >>>>>>>>   explicit flag indicating whether or not extra
> >>>>>>>>
> >>>>>traffic may use the
> >>>>>
> >>>>>>>>   secondary path resources is needed.
> >>>>>>>>
> >>>>>>>??? This is the purpose of this bit.
> >>>>>>>
> >>>>>>This is not clear from the definition.  The bit is defined
> >>>>>>
> >>>>>as indicating
> >>>>>
> >>>>>>the LSP is a secondary (or protecting) LSP and in 1+1
> >>>>>>
> >>>>protection the
> >>>>
> >>>>>>secondary LSP may not be used for extra traffic.
> >>>>>>
> >>>>>>Perhaps the problem here is that protection features are
> >>>>>>
> >>>>>being defined
> >>>>>
> >>>>>>before the protection framework and requirements are
> >>>>>>
> >>>>done.  Is this
> >>>>
> >>>>>>presupposing some particular outcome of the recovery work in CAMP?
> >>>>>>
> >>>>>>JD:  I think the definition of the bit is fine.  For both
> >>>>>>
> >>>>>1+1 and 1:1
> >>>>>
> >>>>>>protection, there would be a pair of Primary LSPs between
> >>>>>>
> >>>>the source
> >>>>
> >>>>>>and destination, rather than a Primary and a Secondary.
> >>>>>>
> >>>>>This is an unusual use of terms.  I have never encountered a case
> >>>>>where both the working and recovery paths are call "primary."
> >>>>>
> >>>>>This is not consistent with either draft-mpls-recovery-framework
> >>>>>or with draft-lang-ccamp-recovery.  I think this is a sign that the
> >>>>>protection work is immature and not ready for progressing to RFC.
> >>>>>
> >>>>>
> >>>>For 1+1 path protection, both working/recovery paths are
> >>>>carrying user data
> >>>>traffic and it is an endpoint decision as to which path is
> >>>>actually the
> >>>>working/recovery path.  At a transit node, both paths need to
> >>>>be treated as
> >>>>primary, as the resources are currently being used and
> >>>>obviously can't be
> >>>>used for Extra Traffic.
> >>>>
> >>>>
> >