[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: RE : RE : TR : I-D ACTION:draft-leroux-ccamp-gmpls-mrn-eval-00.txt

To: dpapadimitriou@psg.com, dimitri.papadimitriou@alcatel.be
Subject: Re: RE : RE : TR : I-D ACTION:draft-leroux-ccamp-gmpls-mrn-eval-00.txt
From: Igor Bryskin <i_bryskin@yahoo.com>
Date: Sat, 5 Mar 2005 10:50:54 -0800 (PST)
Cc: Dimitri.Papadimitriou@alcatel.be, Igor Bryskin <ibryskin@movaz.com>, LE ROUX Jean-Louis RD-CORE-LAN <jeanlouis.leroux@francetelecom.com>, ccamp@ops.ietf.org
Comment: DomainKeys? See http://antispam.yahoo.com/domainkeys
Domainkey-signature: a=rsa-sha1; q=dns; c=nofws; s=s1024; d=yahoo.com; b=y4CYz/xM+wVfrTxVflmwKoigseGLCu509ypm9ibzfdfdhXRr6rk2euW3HFD94AVuWddaqc2EGm430rKJQEG3YIiLcgIviWlSn2kz6iO9v9T/fkX0UvKwqpjuMqos8UHo6dJn4Mou9oJIabNX2dtGD0t6FMR0G7BKl7UvD6Oge3Q= ;
In-reply-to: <4229CA36.6000505@psg.com>

Excellent. Looking forward to the discussion.
Hopefully, I?ll get to understand what is not clean
about the mechanism I suggested.

Before we start the discussion, I?d like to make clear
that all I am asking is a capability to provision
working LSPs as Secondary LSPs. Note that this is a
generic mechanism, and signaling of virtual FAs is one
of possibly many applications that can make use of it.
For example, on many occasions I hear a requirement
from service providers to provide a way of
provisioning a service when part of network resources
are not in place yet with actual service activation
after all resources are installed. This is necessary
to make sure that resources that are in place are not
?grabbed? by other dynamically provisioned services.
This problem could be easily solved by the suggested
mechanism. 

Lastly, despite what you and JL are saying,
provisioning of virtual FAs has a lot to do with
recovery. It is usually a MUST requirement for FAs to
be protected. In this case if you provision virtual FA
as unprotected, and, while it is transformed into real
FA, try to compute/provision protection LSP, the
computation may fail despite that two disjoint paths
do exist and could be determined if they were computed
at the same time. Therefore, it should be possible to
compute working and protection (possibly multiple as
in case of segment protection) path(s) for virtual FA
and signal them, associating them together, without
committing network resource for neither of them.

See you in Minneapolis,
Igor
   


--- dimitri papadimitriou <dpapadimitriou@psg.com>
wrote:

> igor - i will provide a detailed answer concerning a
> clean method to 
> deal with "virtual FAs" that does not rely on the
> use of the protection 
> object (and therefore would deliver a clean method
> to provide signaling 
> for protected "virtual FAs"
> 
> hopefully we can discuss this during this IETF
> meeting,
> 
> thanks,
> - dimitri.
> 
> Igor Bryskin wrote:
> 
> > Dimitri,
> > 
> > See at the bottom
> > 
> > 
> >>>IB>>Well, are they? They give you a way to signal
> a
> >>>requirement not to commit resources for an LSP
> > 
> > until
> > 
> >>>further notice and also to require their
> commitment
> >>>when it becomes necessary. Why not to allow
> working
> >>>LSP to be signaled as Secondary according to the
> >>>mentioned draft? Then it would be possible even
> to
> >>>signal protected virtual FAs, which is not as
> > 
> > stupid
> > 
> >>>as it may sound :=)
> >>
> >>Sorry but this sounds really bad...
> >>Please let me know then how do you signal
> protected
> > 
> > virtual FAs?
> > 
> >>IB>> You signal both working and protection LSPs
> of
> > 
> > the FA as Secondary.
> > At
> > 
> >>the time when you decide to make FA "real" you
> > 
> > re-signal working or both
> > 
> >>LSPs (depending on protection type) as Primary.
> > 
> > Don't see why this
> > wouldn't
> > 
> >>work.
> >>
> >>DP> because what do you do in case the primary
> > 
> > (virtual) FA-LSP fails ?
> > you
> > 
> >>are simply going to activate the secondary
> (virtual)
> > 
> > FA-LSP which is in
> > 
> >>clear
> >>opposition with the notion of virtual-FA - the
> > 
> > reason is again because you
> > 
> >>are trying to use one bit i.e. S to say something
> in
> > 
> > a context (protection
> > 
> >>object) for which it has not been designed for
> >>
> > 
> > 
> > IB>> Oh-Oh, what do you mean by "primary (virtual)
> > FA-LSP fails" and  "you
> > are simply going to activate the secondary
> (virtual)
> > FA-LSP "?  I activate
> > protection LSP if I have a failure in the *data
> plane*
> > of the working LSP.
> > But working LSP of the virtual FA does not have
> any
> > data plane yet, there is
> > nothing that can trigger the protection LSP
> activation
> > and switchover.
> > 
> > DP> but they are associated to resources they do
> not
> > use - and the control plane role is to ensure that
> > when they need to be activated resources can be
> used -
> > the question JL raises is the problem of "how do i
> > protect a virtual FA" and the response is if you
> are
> > using the S bit for telling i am a virtual FA you
> can
> > not use it anymore to ensure protection of a
> primary
> > (and even less a secondary !) - this said an
> secondary
> > LSP exists only within the context of its primary
> > 
> > IB>>Here is what I have in mind.
> > 
> > A network operator configures a mesh of virtual
> FAs.
> > Working FA LSPs are signaled as Secondary. Once
> they
> > are established the FA TE link ends are
> advertised.
> > Resources for these LSPs are allocated on the
> links
> > the LSPs are going through, however, they are not
> > committed (not bound into cross-connects). Two or
> more
> > virtual FA LSPs may share the resources on common
> > links. This is especially true for those of them
> that
> > are unlikely to become real at the same time.
> > Furthermore, polices could be such that virtual FA
> LSP
> > resources could be allowed to get pre-empted by
> real
> > (not necessarily FA-) LSPs. The mentioned
> resources
> > also could be involved in mesh restoration schemes
> and
> > even be bound into extra traffic LSPs. 
> > 
> > When a virtual FA transforms into real FA, its
> working
> > FA-LSP is re-signaled as Primary. At this time,
> the
> > allocated resources are bound into FA-LSP
> > cross-connects. Naturally, all other LSPs ?
> virtual or
> > real ? that share resources with the FA-LSP in
> > question are notified and need to be recomputed
> and/or
> > re-signaled.
> > 
> > A virtual FA could be provisioned as protected in
> a
> > sense that when it becomes real it will have a
> certain
> > protection scheme already in place. Path
> computation
> > for backup LSPs of virtual FAs is performed just
> like
> > of real FAs, that is, by assuming that associated
> > working LSPs are Primary (fully built) LSPs.
> > Regardless of the protection type backup LSPs of
> > virtual FAs are always signaled as Secondary. At
> the
> > time when a virtual FA is transformed into real
> one
> > (that is, when its working LSP becomes Primary)
> its
> > backup LSP(s) may be activated as well (e.g. 1+1
> > protection) or may be left Secondary until the
> working
> > LSP data plane failure is detected (e.g. shared
> > protection).
> > 
> > Igor
> > 
> > 
> > 
> > 
> > 
> > __________________________________________________
> > Do You Yahoo!?
> > Tired of spam?  Yahoo! Mail has the best spam
> protection around 
> > http://mail.yahoo.com 
> > 
> > 
> > .
> > 
> 
>



	
		
__________________________________ 
Celebrate Yahoo!'s 10th Birthday! 
Yahoo! Netrospective: 100 Moments of the Web 
http://birthday.yahoo.com/netrospective/

References:
- Re: RE : RE : TR : I-D ACTION:draft-leroux-ccamp-gmpls-mrn-eval-00.txt
  - From: dimitri papadimitriou <dpapadimitriou@psg.com>

Prev by Date: Re: A philosophical question about FA and FA-LSP
Next by Date: Re: A philosophical question about FA and FA-LSP
Previous by thread: Re: RE : RE : TR : I-D ACTION:draft-leroux-ccamp-gmpls-mrn-eval-00.txt
Next by thread: GMPLS Signaling Procedure for Egress Control
Index(es):
- Date
- Thread