Re: Why we consider the method of "label sharing for fast PE protection"

[Jakob Heitz <[e-mail]>]

I any PE is part of 2 RGs, then those RGs can not have overlapping label spaces. 

--
Jakob Heitz.


On Nov 28, 2013, at 6:16 PM, "Mingui Zhang" <[e-mail]> wrote:

> Hi Bruno, 
> 
> So you suppose the label used in an RG cannot be used again out of the RG. That is not correct.
> Please find my comments inline [Mingui].
> 
> <snip>
>> [Bruno2] Let's assume:
>> - 5 PE in the group, hence sharing the same range of labels (e.g., 1~1100).
>> - 5 VPNs in connect to this group of PE, 2 of which being dual-homed (VPN1 &
>> VPN2).
>> 
>> With label sharing:
>> Label:1    2    3    4    5
>> 
>> PE1    VPN1    x    x    x    x
>> PE2    VPN1    VPN2    x    x    x
>> PE3    x    VPN2    x    x    x
>> PE4    x    x    VPN3    x    x
>> PE5    x    x    x    VPN4    VPN5
>> 
>> All labels marked as "x" are burned/lost because of the label sharing.
> 
> [Mingui] Not true. Where we got this constraint? For an explicitly example, PE4 can well use label 1,2,4,5. 
> 
> [Mingui] I anticipate you assume PE1~PE5 are forming an RG, so that once a label is used it is used across the RG. I need to point out that the unit of "RG" is independent of PEs. It depends on the VPN connections. I saw Zhou Peng has already given examples on this point. 
> 
> <snip>
> 
>> [Bruno2] not always. There is public/ietf example for this:
>> draft-l3vpn-legacy-rtc-00
> 
> [Mingui] It's designed to be incrementally deployable in the network. The trick is confined in the RG. Other P and PE routers are unaware of the change.
> 
> [Mingui] I guess you may change to imagine the scenario that operator need a legacy PE and a label sharing PE form an RG. Let's consider the analogy that the operator interconnects two switches using LAG while one of them does not support LAG at all. :) 
> 
> [Mingui] Thanks for continuing the discussion. I think the discussion about label ranges reservation in another thread is related to our discussion. To my understanding, the conclusion is that it's not OK to require a label block to be supported across multiple PEs. A possible escape is to resort to a higher-layer authorized entity out of the RG to assign the label.
> 
> Thanks,
> Mingui