RE: Questions on draft-ietf-l3vpn-vpn-vr-02.txt

["Vishal Sharma" <v.sharma at ieee.org>]

Title: RE: Questions on draft-ietf-l3vpn-vpn-vr-02.txt

Paul, Benson,

 

Thanks very much for both your clarifications.

 

Given that PE-PE is in fact VR-VR routing, I think MAY is probably the right choice.

 

Although, Benson brings up an excellent point, which I feel would be worth incorporating

in the draft itself.

 

Namely, if the PE-PE (VR-VR) and PE-CE (VR-CE) routing protocol is the same, then it is desirable to have

different processes (for guarding against crashing etc. that Paul mentioned), but at the same time

useful to not have unnecessary re-distribution, info. loss, etc.

 

(Aside: if the two routing protocols are the same, but implemented as separate processes, is

it still possible to avoid redistribution of info. from one to the other??)

 

But, if the PE-CE (VR-CE) links are treated as sources of external reachability for the

PE-PE (VR-VR) domain,

then the two routing protocol implementations MAY be independent.

 

(In other words, it could be two different processes, corresponding to different protocols,

or different instances of the same protocol.)

 

-Vishal

-----Original Message-----
From: Schliesser, Benson [mailto:benson.schliesser at savvis.net]
Sent: Thursday, June 17, 2004 3:01 PM
To: Paul Knight; v.sharma at ieee.org; L3PPVPN
Subject: RE: Questions on draft-ietf-l3vpn-vpn-vr-02.txt

Paul, Vishal-

 

I think it would be good to note (as Vishal suggests) that PE-PE routing is specifically VR-VR routing, typically between PEs. (as differs from the VRF-shared PE-PE routing in BGP/MPLS model)

 

That clarified, I would think that SHOULD is too strong when describing the separation of PE-CE and PE-PE routing protocols. >From a platform implementation perspective, you might say that it's desirable to have separate processes. However, from a network design perspective it is desirable that the SP routing domain is transparent to the customer sites--there is no protocol redistribution, information loss, etc., if the PE-CE routing protocol is the same as the PE-PE routing protocol. (i.e., the VR is almost like a real router...) If being "independent" means that you treat PE-CE links as sources of external reachability for the PE-PE domain, then I'd say MAY is as strong as the draft should require.

 

-Benson

 

---

From: l3vpn-bounces at ietf.org [mailto:l3vpn-bounces at ietf.org] On Behalf Of Paul Knight
Sent: Thursday, June 17, 2004 3:22 PM
To: v.sharma at ieee.org; L3PPVPN
Subject: RE: Questions on draft-ietf-l3vpn-vpn-vr-02.txt

Hi Vishal,

Thanks, these are helpful comments.

In terms of the draft, I guess this boils down to: should we use MUST instead of MAY in this instance?  I think MUST is too strong, without going into a formal definition of "independence."  For instance, if BGP is used for both CE-PE and PE-PE, and BGP crashes on the PE device, are both the PE-PE and CE-PE routing affected?  In some implementations of VR, one or the other BGP session will survive, while in others, they will both fail if BGP crashes.  So even if they are configured as different routing domains, they may not be truly "independent."  However, I would be comfortable with using SHOULD here, i.e.:

"The routing protocol between the CE and the VR running on the PE SHOULD be independent of the PE-to-PE routing."

The original intent of this text in 2.5.1 was to emphasize that the CE-side routing can use a different routing protocol from the core routing.

I agree with your comments on clarifying the CE-to-PE routing and on adding "network" to the backbone VR phrase in 2.5.3.

Thanks and regards,
Paul Knight
Editor, draft-ietf-l3vpn-vpn-vr-02.txt

> -----Original Message-----
> From: Vishal Sharma [mailto:v.sharma at ieee.org]
> Sent: Thursday, June 17, 2004 2:12 PM
> To: L3PPVPN; Knight, Paul [BL60:1A14:EXCH]
> Subject: RE: Questions on draft-ietf-l3vpn-vpn-vr-02.txt
>
>
> Paul and co.,
>
> Perhaps I should clarify one of my questions...
>
> With respect to the independence of the CE-PE routing
> protocol from the PE-PE routing protocol, when you said
> "PE-to-PE routing protocol" in 2.5.1. were you referring to
> the routing protocol running between the VRs (on the PEs)
> corresponding to the VPN of which the CE is a part?
>
> If so, then I can see that these two protocol MAY be independent.
>
> If not, then I am confused, because my understadning is
> that the true PE-to-PE routing protocol
> (presumably the IGP instance which lets all PEs and Ps in the
> SP network learn reachability to one another) should _always_
> be independent of the CE-to-PE (read CE-to-corresponding VR
> on attached PE) routing protocol.
>
> Thanks,
> -Vishal
>
> > -----Original Message-----
> > From: Vishal Sharma [mailto:v.sharma at ieee.org]
> > Sent: Thursday, June 17, 2004 10:58 AM
> > To: L3PPVPN; Paul Knight
> > Subject: Questions on draft-ietf-l3vpn-vpn-vr-02.txt
> >
> >
> > Hi Paul and co-authors,
> >
> > Could you please clarify a few things in the latest version of this
> > document:
> >
> > - Sec. 2.5.1. states that the
> > "The routing
> >    protocol between the CE and the VR running on the PE MAY be
> >    independent of the PE-to-PE routing. "
> >
> > Why is this a MAY? Would one not want the routing protocol
> between the
> > CE and its corresponding VR on the PE to be always
> independent of the
> > PE-to-PE routing protocol? When is there a case when these
> protocols
> > are not independent?
> >
> > - Sec. 2.5.3, I assume when "CE-to-PE routing" is used, it actually
> > means "routing between the CE and the VR (on the PE to
> which the CE is
> > attached) corresponding to the VPN of which the CE is a
> part" Is this
> > correct?
> >
> > If so, it might be useful for the document to clarify somewhere up
> > front that when "CE-to-PE" routing is used, this is what is implied.
> >
> > - Sec. 2.5.3, "VRs
> >    of customer VPNs MAY use the backbone VR for routing across the
> >    backbone." probably needs to say "the backbone VR network".
> >
> > Thanks,
> > -Vishal
> >
>
>