Re: [Softwires] BGP TE attr last call by softwires WG (2nd question)

[Igor Bryskin <[e-mail]>]

Hi John,

I understand what you are saying and disagree. The overlay I am talking about logically is a separate network and as any network it should be sufficiently redundant to function. There is a number of ways how you can address the redundancy concerns. Look at the examples below:

a) interconnect all VPN-aware PEs into a single ring: 
PE=======PE
 ||                  ||
PE               PE 
 ||                   ||
PE               PE 
||                   ||
...                ....
PE=======PE

b) connect each PE to two interconnected Ps 

PE               P                 PE
                    ||                   
PE               ||                  PE 
                    ||                  
PE               ||                  PE 
                    ||                  
...                 ||                   ....
PE               P                 PE


Note that tunnels can traverse any number of VPN-unaware Ps and PEs.

Igor



----- Original Message ----
From: "Drake, John E" <[e-mail]>
To: Igor Bryskin <[e-mail]>; Yakov Rekhter <[e-mail]>; Lou Berger <[e-mail]>
Cc: Yakov Rekhter <[e-mail]>; Adrian Farrel <[e-mail]>; [e-mail]; [e-mail]
Sent: Tuesday, September 2, 2008 2:24:26 PM
Subject: RE: [Softwires] BGP TE attr last call by softwires WG (2nd question)

Igor,

Several years ago when OSPF was first proposed as an autodiscovery
mechanism for L1VPNs, you were told that it was a bad idea due to its
scaling properties and impact on the IGP.

You are now tacitly agreeing with those who told you it was a bad idea.

For your suggested approach to work with sufficient redundancy, the
topology of the overlay needs to be configured such that every selected
P router is connected to at least two other selected P routers and every
PE router needs to be connected to at least two selected P routers.

When you are done with this configuration, you are left with a situation
in which *every* PE and selected P router will have *all* L1VPN routes.

Thanks,

John 

>-----Original Message-----
>From: Igor Bryskin [mailto:[e-mail]] 
>Sent: Friday, August 29, 2008 12:10 PM
>To: Drake, John E; Yakov Rekhter; Lou Berger
>Cc: Yakov Rekhter; Adrian Farrel; [e-mail]; 
>[e-mail]
>Subject: Re: [Softwires] BGP TE attr last call by softwires WG 
>(2nd question)
>
>Are you calling me silly? Are you coming to Minneapolis? :=)
>
>Seriously, what is wrong in your opinion with this approach? 
>Many people are talking about multi-instance IGPs. What they 
>have in mind is improving the IGP scalability:
>a) by removing non-IP advertisements from the instance of IGP 
>that manages IP routing/forwarding tables into separate IGP 
>instance(s);
>b) by distributing non-IP information only to and via 
>inerested parties leaving the bulk of Ps out of the process.
>
>In my opinion this is exactly what is needed for the 
>OSPF-based L1VPN application.
>
>Igor
>
>
>
>
>
>
>----- Original Message ----
>From: "Drake, John E" <[e-mail]>
>To: Igor Bryskin <[e-mail]>; Yakov Rekhter 
><[e-mail]>; Lou Berger <[e-mail]>
>Cc: Yakov Rekhter <[e-mail]>; Adrian Farrel 
><[e-mail]>; [e-mail]; [e-mail]
>Sent: Friday, August 29, 2008 2:31:36 PM
>Subject: RE: [Softwires] BGP TE attr last call by softwires WG 
>(2nd question)
>
>So you are proposing an OSPF route reflector?  At what point 
>does the silliness stop? 
>
>>-----Original Message-----
>>From: Igor Bryskin [mailto:[e-mail]]
>>Sent: Friday, August 29, 2008 11:29 AM
>>To: Drake, John E; Yakov Rekhter; Lou Berger
>>Cc: Yakov Rekhter; Adrian Farrel; [e-mail]; 
>>[e-mail]
>>Subject: Re: [Softwires] BGP TE attr last call by softwires WG (2nd 
>>question)
>>
>>Hi John,
>>
>>No, not really. When you add a PE you configure local 
>interfaces, local 
>>VPN port mappings, stuff like that. While doing this you will also 
>>configure an IPinIP tunnel to one of your spoke Ps and enable L1VPN 
>>OSPF instance on the tunnel.
>>Once you did that the local VPN information will be flooded 
>accross the 
>>overlay, likewise, the new PE will get all the necessary information 
>>from other PEs.
>>
>>Cheers,
>>Igor
>>
>>
>>----- Original Message ----
>>From: "Drake, John E" <[e-mail]>
>>To: Igor Bryskin <[e-mail]>; Yakov Rekhter 
>><[e-mail]>; Lou Berger <[e-mail]>
>>Cc: Yakov Rekhter <[e-mail]>; Adrian Farrel 
>><[e-mail]>; [e-mail]; [e-mail]
>>Sent: Friday, August 29, 2008 11:20:16 AM
>>Subject: RE: [Softwires] BGP TE attr last call by softwires WG (2nd 
>>question)
>>
>>Igor,
>>
>>Doesn't this defeat auto-discovery?  I.e., how is a new PE added to a 
>>given L1VPN?
>>
>>Thanks,
>>
>>John
>>
>>>-----Original Message-----
>>>From: Igor Bryskin [mailto:[e-mail]]
>>>Sent: Friday, August 29, 2008 5:51 AM
>>>To: Yakov Rekhter; Lou Berger
>>>Cc: Yakov Rekhter; Adrian Farrel; [e-mail]; 
>>>[e-mail]
>>>Subject: Re: [Softwires] BGP TE attr last call by softwires WG (2nd
>>>question)
>>>
>>>Yakov,
>>>
>>>You said:
>>>
>>>
>>>... And while on the subject of scaling, please keep in mind 
>that BGP 
>>>only stores L1VPN routes on PEs that have sites of that VPN 
>connected 
>>>to them, and on an RR if used, but *not* on any of the P routers. In 
>>>contrast, rfc5252 (OSPF for L1VPN
>>>autodiscovery) results in storing *all VPN TE information for all the
>>>VPNs* on *all* the IGP nodes, both P and PE. So, clearly BGP-based 
>>>approach scales better than OSPF-based approach.
>>>
>>>Yakov.
>>>
>>>This is not true in case of multi-instance OSPF: one can build an 
>>>overlay interconnecting PEs via one or small number of Ps
>>using IPinIP
>>>tunnels and run in this overlay an instance of OSPF specifically 
>>>designated for distribution of L1VPN information. In this
>>case the OSPF
>>>solution won't scale any worse than the BGP approach. Note. 
>>that rfc252
>>>never said that the instance of OSPF used for flooding of the L1VPN 
>>>information must be the same instance that is used for the
>>distribution
>>>of IP-related LSAs.
>>>
>>>Regards,
>>>Igor
>>>
>>>
>>>
>>>
>>
>>
>>
>>
>
>
>