[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
RE: [PWE3] PWE3 question
Sasha,
> -----Original Message-----
> From: Sasha Vainshtein [mailto:Sasha@axerra.com]
> Sent: Wednesday, May 15, 2002 10:11 AM
> To: 'Rutemiller, John'
> Cc: 'pwe3@ietf.org'
> Subject: RE: [PWE3] PWE3 question
>
>
> John,
> Please see some comments/responses inline.
> Hopefully they will clarify my position.
> With best regards,
> Sasha Vainshtein
> email: sasha@axerra.com <mailto:sasha@axerra.com>
> tel: +972-3-7659993 (office)
> +972-8-9254948 (res.)
> +972-58-674833 (cell.)
>
>
>
> > -----Original Message-----
> > From: Rutemiller, John [mailto:John.Rutemiller@marconi.com]
> > Sent: Wednesday, May 15, 2002 12:12 PM
> > To: 'pwe3@ietf.org'
> > Subject: RE: [PWE3] PWE3 question
> >
> >
> >
> >
> > > -----Original Message-----
> > > From: Sasha Vainshtein [mailto:Sasha@AXERRA.com]
> > > Sent: Friday, May 10, 2002 2:57 AM
> > > To: 'Shahram Davari'
> > > Cc: 'pwe3@ietf.org'; Danny McPherson (E-mail); 'Andrew G.
> > > Malis'; Neil.
> > > 2. Harrison (E-mail)
> > > Subject: RE: [PWE3] PWE3 question
> > >
> > >
> > > Shahram and all.
> > > Please see a brief comment inline.
> > >
> > [SNIP]
> > >
> > > I would suggest a very simple criterion regarding exertion or
> > > non-exertion of control over PSN: if setup/teradown of a PW
> > > requires modification of the FIB in at least one PSN node that
> > > does not act as the PE for this PW, it means that control over
> > > this router has been exerted.
> > >
> > > This criterion does not depend on the protocol being used
> > > (static,LDP, RSVP or else).
> > >
> > > The single-label model requires allocation of a new label per PW
> > > in each interior node on the path fromingress to egress, and hence
> > > means exertion of control over the PSN unless the the pair of
> > > PEs are adjacent (the "single hop" case mentioned by Neil).
> > >
> > > Can we agree on that?
> > >
> >
> >
> > Creation of a Pseudo wire requires a tunnel label.
> >
> Correct. You must associate every direction of a PW
> over an MPLS network with an LSP leading from the
> ingress PE (for this direction) to egress PE.
> >
> >The creation of this tunnel must be configured.
> >
> It can be created regardless of any specific PW that would use it.
But if you don't have a PW between two PEs, you may not need any
tunnel between them.
> And it could be created in many ways, including hop-by-hop forwarding
> provided by the vanilla LDP.
> Once created, such a tunnel can be used by - and, for scalability
> reasons, SHOULD be used - by more than one PW between the
> same pair of PEs.
>
> And it could be created in many ways, including "vanilla" LDP
> in DU mode.
> >
> >The establishment of the tunnel will affect the FIB if the P
> routers.
> >
> Correct. However, as I said above, creation of the tunnel (to be used
> by multiple PWs) is not caused by creation of the PWs. Creation
> of PWs using an already existing tunnel would not affect FIB in the
> P routers, only in PEs.
But what if a tunnel does not already exist? Then you need to create
a tunnel for the first PW between two PEs.
> >
> >Furthermore, the tunnel must be created with a forwarding class
> >that is sufficient to carry the desired pseudo wire traffic.
> >
> There are many ways to associate a certain forwarding behavior
> with the EXP bits across an LSP. E.g., one can use
> Preconfigured EXP<-->PHB Mapping in LSPs that have been
> created by LDP in DU mode.
>
> Such an method cannot be used if reservation of resources
> must be associated with an LSP. But even then nothing
> prevents creation of a "planned" network of DiffServ-enabled
> LSPs with BW reservation beforehand regardless of any
> specific PW, and then creation of PWs that use available
> capaity on this "network".
Sure, but the path that the tunnel should take depends on the
traffic parameters of the PWs inside that tunnel. In other words
the tunnel path computation needs to take into account the BW
requirement of the PWs, and if the BW of an existing tunnel is not
enough and can't be increased, then another tunnel needs to be created.
> >
> > If traffic classes are to be separated (e.g, ATM CBR gets EF and
> > ATM UBR gets best effort), then multiple tunnels must be created
> > across the network.
> >
> Not necessarily. A single E-LSP will may support up to 8 different
> traffic classes. And this can be done by using preconfigured
> EXP<-->PHB mappings, too.
Sure, but you can't do TE with E-LSPs ! (See DS-aware TE drafts)
> >
> >
> > Therefore, the PW has in some way exerted control over the P routers
> > even with the tunnel creation.
> >
> This would only happen if you apply the "One PW = One Tunnel" model
No, even in case of N PW = 1 tunnel you still need to create the tunnel.
Even if a tunnel already exist between two PEs, it may not be a suitable
tunnel for carrying the PWs (e.g., may not have enough BW).
> that originated this thread. And that is exactly why I said
> that, IMHO,
> this model is beyound the chartered PWE3 scope. The model that
> separates creation of transport tunnels from creation of PWs is
> within this scope.
I think what John was trying to say is that, Even in case of tunneled
PWs, the P routers' configuration and FIB tables are not 100% PW independent.
Yours,
-Shahram
> >
> > John
> >
> > _______________________________________________
> > pwe3 mailing list
> > pwe3@ietf.org
> > https://www1.ietf.org/mailman/listinfo/pwe3
> >
>
> _______________________________________________
> pwe3 mailing list
> pwe3@ietf.org
> https://www1.ietf.org/mailman/listinfo/pwe3
>
_______________________________________________
pwe3 mailing list
pwe3@ietf.org
https://www1.ietf.org/mailman/listinfo/pwe3