Re: [bess] A question about the draft-wang-bess-sbfd-discriminator

[Greg Mirsky <[e-mail]>]

Hi Haibo,
thank you for the clarification. I may suggest a text for Section 3:

In some EVPN deployments, for example, when it spans over multiple domains,
only one of a pair of interconnected PEs benefits from monitoring the
status of the connection. In such a case, using S-BFD [RFC7880] is
advantageous as it reduces the load on one of the PEs while providing the
benefit of faster convergence. The following sections provide examples of
EVPNs that would benefit from using S-BFD.

What are your thoughts?

Regards,
Greg

On Tue, Mar 15, 2022 at 7:18 PM Wanghaibo (Rainsword) <
[e-mail]> wrote:

> Hi Greg,
>
>        Thanks for you comments.
>
> Yes, the resources will save at PE1 and PE2 as figure 1. This is a typical
> 3PE scenario.
>
>        The service is like this:
>
> +-----+    +-----+        +-----+
>
> | UCE1|----| APE1|--------|SPE1 |,
>
> +-----+    +-----+`      /+-----+ `.
>
>                    `,  .'           `.+-----+
>
>            ......    \/               | SCE1|
>
>                      /\              `+-----+
>
>                     `  `.          ,'
>
> +-----+    +-----+,'     .+-----+ `
>
> | uCEn|----| APEn|--------|SPE2 |`
>
> +-----+    +-----+        +-----+
>
>        There may be many Access PEs，used to access User CE. And they all
> multi-homed to a couple Servicc PE, SPE1 and SPE2.
>
>        (shown as the PE1 and PE2 as figure 1)
>
>         Access PE needs to detect Service PE’s reachability. Access PE
> creates SBFD session as an initiator, SPE as the reflector. This will save
> Service PEs’ resources.
>
>
>
> Regards,
>
> Haibo
>
>
>
> *From:* Greg Mirsky [mailto:[e-mail]]
> *Sent:* Tuesday, March 15, 2022 11:12 PM
> *To:* Wanghaibo (Rainsword) <[e-mail]>
> *Cc:* [e-mail]; BESS <[e-mail]>;
> rtg-bfd WG <[e-mail]>
> *Subject:* Re: A question about the draft-wang-bess-sbfd-discriminator
>
>
>
> Hi Haibo,
>
> thank you for your expedient response. If I understand the scenario you're
> addressing, it is where a single PE with moderate resources is connected to
> a PE that acts as the edge device for the access network. To improve the
> quality of user experience, customer's PE is connected to a secondary PE
> that is used as a backup. You are concerned that maintaining two BFD
> sessions on the customer's PE might overload the resource-limited PE. But
> isn't that the PE that initiates S-BFD sessions toward two access
> network edge PEs in your draft? I think that the savings are on the side of
> these two PEs, not the subscriber's PE. Would you agree?
>
>
>
> Regards,
>
> Greg
>
>
>
> On Tue, Mar 15, 2022 at 7:20 AM Wanghaibo (Rainsword) <
> [e-mail]> wrote:
>
> Hi Greg,
>
>        Thanks for your comments.
>
>        The scenario you pointed out is a 4PE scenario, but in our
> solution, a large number of scenarios are based on 3PE.
>
> In a 3PE scenario, deploying BFD wastes resources. A large number of
> single-homed PEs may be connected to the dual-homed PEs. The dual-homed PEs
> may not have enough resources to create BFD sessions.
>
>
>
> Regards,
>
> Haibo
>
>
>
> *From:* Greg Mirsky [mailto:[e-mail]]
> *Sent:* Tuesday, March 15, 2022 12:44 AM
> *To:* Wanghaibo (Rainsword) <[e-mail]>;
> [e-mail]; BESS <[e-mail]>;
> rtg-bfd WG <[e-mail]>
> *Subject:* A question about the draft-wang-bess-sbfd-discriminator
>
>
>
> Hi Haibo and the Authors,
>
> thank you for updating the draft. I've read the new version and have a
> question about the use case presented in the document. There are three PEs
> with two of them providing redundant access to a CE. It appears that a more
> general case would be if both CEs use redundant connections to the EVPN.
> Asume, PE4 is added and connected to CE2. In that case, it seems reasonable
> that each PE is monitoring remote PEs, i.e., PE1 monitors PE3 and PE4, PE2
> - PE3 and PE4, PE3 - PE1 and PE2, and PE4 - PE1 and PE2. So, now there are
> pairs of S-BFD sessions between PEs connected to CE1 and CE2 respectively.
> That seems like too many sessions and that number can be reduced if one
> uses BFD instead of S-BFD. Would you agree? To simplify operations, it
> might be helpful to use the technique described in
> draft-ietf-bfd-unsolicited
> <https://datatracker.ietf.org/doc/html/draft-ietf-bfd-unsolicited-09>. In
> the recent discussion of the draft on the BFD WG ML, the authors noted that
> they are working on extending the scope to include the multi-hop BFD.
>
> Greatly appreciate your thoughts about the number of S-BFD sessions.
>
>
>
> Regards,
>
> Greg
>
>