Re: [Pce] PCEP ERO

[Igor Bryskin <[e-mail]>]

Eric,
Note that this discussion is not about LOOSE flag or where the path computation takes place. It is about which interfaces wrt source->destination path direction are allowed to be put into ERO:
a) only node IDs and inbound interface IDs (as the mentioned RFC recommends);
b) only node IDs and outbound interface IDs (as some implementations I am aware of do);
c) both inbound and outbound interface IDs;
d) anything - an arbitrary set of either node, inbound or/and outbound interface IDs.

Note also, that we are talking not only about RSVP EROs, but also about PCEP EROs, for example, an ERO that allows for a PCC to express the "inclusion" constraint - an ordered list of nodes, interfaces, labels, etc. that must appear in the resulting path in the specified order, with the path computation "filling the gaps" in front of specified loose elements.

My suggestion is d). This would provide a lot of flexibility and can be easily disambiguated by introducing an additional ERSO flag - "OUT", which should be set for numbered or unnumbered ERSOs associated with outbound interface IDs.

Having said that, please, see my answers in-line.

Cheers,
Igor

-----Original Message-----
From: Eric Gray [mailto:[e-mail]] 
Sent: Wednesday, June 18, 2014 5:08 PM
To: Igor Bryskin; [e-mail]; 'Zhangxian (Xian)'; 'Dhruv Dhody'; 'Julien Meuric'
Cc: [e-mail]
Subject: RE: [Pce] PCEP ERO

Igor,

	Reasonable people can always disagree.  :-)
IB>> Agree, it's been always a pleasure disagreeing with you :-)

	It may very well be the case that the scenario you're describing lacks the element of flexibility required for local decisions to be useful.

	I can see computation by a PCE as being much more scalable than requiring each potential ingress to know about the capabilities of each potential egress, but - in this case - isn't it possible to either push the entire path (for at least a specific domain) from the PCE to the ingress, or use delegation (either approach effectively relieves the ingress from having to "know" anything about the egress)?
IB>> My point is that whether path computation is performed on an ingress NE or a PCE, it should be completed in its entirety in one place (i.e. logically centralized), and produce no loose elements in the resulting paths. Note that I am talking only about WDM layer, in all other layers it is perfectly OK to compute a service path in a distributed way via signaling EROs with loose elements.

	Or is this what you are trying to get at?  If so, and there is really no advantage to local decision-making, what would be the point of using a "loose" ERO?
IB>> See above

--
Eric

-----Original Message-----
From: Igor Bryskin [mailto:[e-mail]]
Sent: Wednesday, June 18, 2014 10:16 AM
To: Eric Gray; [e-mail]; 'Zhangxian (Xian)'; 'Dhruv Dhody'; 'Julien Meuric'
Cc: [e-mail]
Subject: RE: [Pce] PCEP ERO
Importance: High

Hi Eric,
Sorry to say this, but in the context of WDM layer I cannot disagree with you more.
It is my conviction that WDM layer path (considering all the WDM related constraints, such as optical impairments) the entire LSP path in all its details must be computed in one place (ingress NE or PCE) and no flexibility must be left for decisions on transit nodes.

Regards,
Igor

-----Original Message-----
From: Eric Gray [mailto:[e-mail]]
Sent: Wednesday, June 18, 2014 9:46 AM
To: Igor Bryskin; [e-mail]; 'Zhangxian (Xian)'; 'Dhruv Dhody'; 'Julien Meuric'
Cc: [e-mail]
Subject: RE: [Pce] PCEP ERO

Igor,

	I very much dislike the idea of a path ingress, or a point along the path, making a decision as to what egress interface I will use at any later point in the path (path egress, domain egress or elsewhere), based on an awareness that the decision point supposedly has about the capabilities of a downstream node.

	Wouldn't it be better to include the desired capabilities in signaling, and let the downstream node(s) make this decision for themselves?  That way, it is only necessary for a potential decision point in the network to know that some subsequent (downstream) node has this capability without needing to know which interface(s) (or even necessarily which node) this capability is supported via.

	This seems a generally desirable behavior associated with information hiding.  What is different in this case?

--
Eric

-----Original Message-----
From: Pce [mailto:[e-mail]] On Behalf Of Igor Bryskin
Sent: Wednesday, June 18, 2014 7:50 AM
To: [e-mail]; 'Zhangxian (Xian)'; 'Dhruv Dhody'; 'Julien Meuric'
Cc: [e-mail]
Subject: Re: [Pce] PCEP ERO

Adrian,
Please see in-line.

-----Original Message-----
From: Adrian Farrel [mailto:[e-mail]]
Sent: Wednesday, June 18, 2014 5:14 AM
To: Igor Bryskin; 'Zhangxian (Xian)'; 'Dhruv Dhody'; 'Julien Meuric'
Cc: [e-mail]
Subject: RE: [Pce] PCEP ERO

Hi Igor,

Why don't you put the address of NE X in the ERO? You wouldn't even need to say loose (unless it was genuinely a loose hop).

IB>> I am talking about the "loose" scenario. Yes, it is possible to insert address of NE X as loose hop, but this would at least unnecessarily increase the length of the ERO ( loose node ERSO + strict IN interface ERSO vs. loose OUT interface ERSO). More importantly, sometimes I really need to specify OUT interface in the ERO. Consider, for example, the situation where OUT interface Y can provide the optical signal regeneration I need. The way I constrain the path computation to use the needed re-generation is by specifying OUT interface Y ERSO followed by the label ERSO that encodes the necessary regenerator. I cannot do this by using the peer IN part of the link because the regenerator belongs to the interface Y.

Now, maybe you are concerned that, having decided which i/f you want to exit by, there is a risk that you would reach NE X by that interface. 

IB>> No, this is not my concern.

Well, that is ever
so slightly possible, but:

- It seems incredibly unlikely that the out interface you are so determined to use could accidentally become the in interface. If that happened then you would already have reached the downstream node that you wanted to reach. Uck. Sounds like a bit of bad path computation!

- You could use path exclusions to make sure that doesn't happen.

BTW...
Is it time to move this discussion to CCAMP where signaling people can discuss what you are suggesting?

IB>> Well, IMO this is really a path computation problem, more specifically, path computation constraint problem, unless you believe that paths in the WDM layer could be computed in a distributed way, which I don't.

Cheers,
Igor

Adrian

> -----Original Message-----
> From: Igor Bryskin [mailto:[e-mail]]
> Sent: 17 June 2014 13:11
> To: Zhangxian (Xian); [e-mail]; 'Dhruv Dhody'; 'Julien Meuric'
> Cc: [e-mail]
> Subject: RE: [Pce] PCEP ERO
> 
> Hi Xian,
> 
> Here is my problem. What if I want my path to go through NE X  and 
> exit it via interface Y but do not care on which interface the path enters the NE X?
> What I was saying is that just like we have the LOOSE flag, we 
> could've had
"OUT"
> flag, which would solve all kinds of ambiguities.
> 
> Igor
> 
> -----Original Message-----
> From: Zhangxian (Xian) [mailto:[e-mail]]
> Sent: Monday, June 16, 2014 10:17 PM
> To: Igor Bryskin; [e-mail]; 'Dhruv Dhody'; 'Julien Meuric'
> Cc: [e-mail]
> Subject: RE: [Pce] PCEP ERO
> 
> Hi, Igor,
> 
>   I think you raise up a good question.
> 
>   Just wonder if the text in Section 6.1.2 of RFC4990 (copied below) 
> touch
upon
> the very same problem and provide some guidance?
> 
> --------Section 6.1.2 of RFC4990
> There are two differences between Loose and Strict subobjects.
> 
>    o  A subobject marked as a loose hop in an ERO must not be followed
>       by a subobject indicating a label value [RFC3473].
> 
>    o  A subobject marked as a loose hop in an ERO should never include
>       an identifier (i.e., address or ID) of the outgoing interface.
> 
>    There is no way to specify in an ERO whether a subobject identifies
>    an incoming or outgoing TE link.  Path computation must be performed
>    by an LSR when it encounters a loose hop in order to resolve the LSP
>    route to the identified next hop.  If an interface is specified as a
>    loose hop and is treated as an incoming interface, the path
>    computation will select a path that enters an LSR through that
>    interface.  If the interface was intended to be used as an outgoing
>    interface, the computed path may be unsatisfactory and the explicit
>    route in the ERO may be impossible to resolve.  Thus a loose hop that
>    identifies an interface should always identify the incoming TE link
>    in the data plane.
> -----------------
> 
> Regards,
> Xian
> 
> 
> 
> -----Original Message-----
> From: Pce [mailto:[e-mail]] On Behalf Of Igor Bryskin
> Sent: 2014年6月16日 20:23
> To: [e-mail]; 'Dhruv Dhody'; 'Julien Meuric'
> Cc: [e-mail]
> Subject: Re: [Pce] PCEP ERO
> 
> Hello,
> You are right Julien and Adrian, this is a very old issue. One thing 
> that has
been
> missing IMO is a flag in a numbered/unnumbered link ERSO indicating 
> whether the indicated side of the link is outbound or inbound wrt the 
> path direction
from
> its source to destination. The lack of said flag has been especially a 
> problem
when
> combined with the LOOSE flag. Consider, for example, the situation 
> when a 1.1.24.1/loose is found in the ERO. If 1.1.24.1 interface is 
> meant to be
inbound,
> the path should *enter* the NE that terminates the interface. However, 
> If
> 1.1.24.1 interface is meant to be outbound, the path should *exit*the said NE.
> So, if the ERO is specified as a path computation constraint, the PCE 
> may
produce
> very different resulting paths depending on the PCE's assumptions/ 
> interpretations. The introduction of said flag would resolve the 
> ambiguity and provide the flexibility (e.g. Druv is talking about) for the ERO encoding.
> 
> Regards,
> Igor
> 
> 
> -----Original Message-----
> From: Pce [mailto:[e-mail]] On Behalf Of Adrian Farrel
> Sent: Monday, June 16, 2014 7:33 AM
> To: 'Dhruv Dhody'; 'Julien Meuric'
> Cc: [e-mail]
> Subject: Re: [Pce] PCEP ERO
> 
> Julien is right (of course).
> 
> This survey led (in part) to RFC 4990. section 6 may be what Dhruv is 
> looking
for.
> 
> A nasty question lurking in the background is whether a PCC needs to 
> indicate which construction of ERO is prefers. Consider if the 
> interface was CLI not
> PCEP: in this case the supported construction of ERO is part of the 
> CLI
definition.
> However, given that most of the ERO is not for local consumption and 
> does not need to be examined by the PCC, this question may be of debatable value.
> 
> Adrian
> 
> > -----Original Message-----
> > From: Pce [mailto:[e-mail]] On Behalf Of Dhruv Dhody
> > Sent: 16 June 2014 10:27
> > To: Julien Meuric
> > Cc: [e-mail]
> > Subject: Re: [Pce] PCEP ERO
> >
> > Hi Julien,
> >
> > Thanks for the pointer, this surely helps.
> > Time to dive into the archives.....
> >
> > Dhruv
> >
> > On Mon, Jun 16, 2014 at 2:29 PM, Julien Meuric 
> > <[e-mail]>
> > wrote:
> > > Hi Dhruv.
> > >
> > > PCEP does not mandates more rules on ERO than RSVP-TE, which 
> > > reminds me
> > of
> > > an old discussion in CCAMP. You may want to have a look at
> > > http://tools.ietf.org/html/draft-farrel-ccamp-ero-survey-00 and 
> > > dive into the associated thread back in 2006.
> > >
> > > Julien
> > >
> > >
> > > Jun. 16, 2014 - Dhruv Dhody:
> > >>
> > >> Attaching the figure in a pdf, in case you could not view in my 
> > >> previous mail.
> > >>
> > >> Regards,
> > >>
> > >> Dhruv
> > >>
> > >> ---------------------------------------------------------------
> > >>
> > >> *Dhruv Dhody *
> > >>
> > >>
> > >> System Architect,
> > >>
> > >> Huawei Technologies India Pvt. Ltd.,
> > >>
> > >> Banagalore
> > >>
> > >> Mobile: [phone]
> > >>
> > >> This e-mail and its attachments contain confidential information 
> > >> from HUAWEI, which
> > >>
> > >> is intended only for the person or entity whose address is listed above.
> > >> Any use of the
> > >>
> > >> information contained herein in any way (including, but not 
> > >> limited to, total or partial
> > >>
> > >> disclosure, reproduction, or dissemination) by persons other than 
> > >> the intended
> > >>
> > >> recipient(s) is prohibited. If you receive this e-mail in error, 
> > >> please notify the sender by
> > >>
> > >> phone or email immediately and delete it!
> > >>
> > >> *From:*Dhruv Dhody
> > >> *Sent:* 16 June 2014 11:52
> > >> *To:* [e-mail]
> > >> *Subject:* PCEP ERO
> > >>
> > >>
> > >> Dear WG,
> > >>
> > >>
> > >> Consider the below topology, PCE computes a path from RTA to RTC.
> > >>
> > >> This path maybe encoded in PCEP ERO as  -
> > >>
> > >> ~ (10.1.1.1, 10.1.1.2, 20.1.1.1, 20.1.1.2)
> > >>
> > >> or
> > >>
> > >> ~ (10.1.1.2, 20.1.1.1, 20.1.1.2) [without local IP address of 
> > >> ingress]
> > >>
> > >> IMO both should be considered as viable options.
> > >>
> > >> Is there any reason for PCC to consider one of them as incorrect?
> > >>
> > >> Regards,
> > >>
> > >> Dhruv
> > >>
> > >> ---------------------------------------------------------------
> > >>
> > >> Dhruv Dhody
> > >>
> > >> System Architect,
> > >>
> > >> Huawei Technologies India Pvt. Ltd.,
> > >>
> > >> Banagalore
> > >>
> > >> Mobile: [phone]
> > >>
> > >>
> > >>
> > >> _______________________________________________
> > >> Pce mailing list
> > >> [e-mail]
> > >> https://www.ietf.org/mailman/listinfo/pce
> > >>
> > >
> > > _______________________________________________
> > > Pce mailing list
> > > [e-mail]
> > > https://www.ietf.org/mailman/listinfo/pce
> >
> > _______________________________________________
> > Pce mailing list
> > [e-mail]
> > https://www.ietf.org/mailman/listinfo/pce
> 
> _______________________________________________
> Pce mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/pce
> 
> _______________________________________________
> Pce mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/pce

_______________________________________________
Pce mailing list
[e-mail]
https://www.ietf.org/mailman/listinfo/pce