Re: [trill] 答复: Re: Comments on draft-dunbar-trill-directory-assisted-encap

[Donald Eastlake <[e-mail]>]

Hi Fangwei,

On Wed, Aug 21, 2013 at 10:33 PM, <[e-mail]> wrote:
>
>
> Hi, Donald
>
> Please see my comments inline with [hfw]
>
> (1) The remote egress RBridge could not learn the {nickname, MAC} pair by the data frame.
>
> [Linda] The “nickname” is Ingress’ nickname, the MAC-SA is ingress’ too. So the egress RBridge should be able to learn the mapping between the Ingress’ nickname and its MAC. Why do you say “could not”?
>
> (dee3) Right, its asymmetric. So the end station does the ingress but the egress RBridge does the usual egress process, learning the remote MAC address as attached to the ingress nickname in the TRILL Data frame -- but it does not matter that much if the egress RBridge learns anything if you do ingress for traffic going the other way at the end station also.
>
> [hfw] This is one of the scenarios : local end station can do TRILL encapsulation, while the remote endstation only process the native frame.
>
>
> (2) The transit RBridges could not pass the RPF check for multi-destination forwarding.
>
> [Linda] Why? The data frame is same as if the frame is encapsulated by the Ingress RBridge.
>
>  (dee3) The only place where there might be an RPF check problem is on receipt at the first edge RBridge so some care may be needed there. Alternatively, you could just not do the ingress at the end station for multi-destination frames but have them ingressed by the edge RBridge.
>
> [hfw] Is phantom nickname golabl unique or RB unique? For global unique, all the edge RBridges have the same value. In this situation, the RPF does not pass for multi-destination.

If you use a phantom nickname, I think there needs to be one phantom
nickname unique across the campus for each edge RBridge to which
pre-ingressed frames are being sent. The idea is that the edge RBridge
doesn't need to do any different processing (other than accepting
pre-ingressed packets). If you just had one phantom nickname for the
campus, lots of things would not work, such as unicast address
learning. Probably the phantom nickname needs to be reported in LSP as
being adjacent to its associated edge RBridge with zero cost...

>      For RB unique, different edge RBridges have different phantom nicknames. In this situation, it wastes many nicknames. We can assign the RB1's nickname (N1, not phantom nickname) to the non-RBridge nodes.

You can use RB1's nickname if RB1 will accept pre-ingressed frames
where the ingress nickname field is equal to RB1's nickname.

Thanks,
Donald
=============================
 Donald E. Eastlake 3rd   [phone] (cell)
 155 Beaver Street, Milford, MA 01757 USA
 [e-mail]

> Thanks!
>
> Fangwei
>
>
>
> Donald Eastlake <[e-mail]>
> 发件人:  [e-mail]
>
> 2013-08-21 08:46
>
> 收件人
> Linda Dunbar <[e-mail]>
> 抄送
> "[e-mail]" <[e-mail]>rg>, "[e-mail]" <[e-mail]>cn>, "[e-mail]" <[e-mail]>
> 主题
> Re: [trill] Comments on draft-dunbar-trill-directory-assisted-encap
>
>
>
>
>
> Hi,
>
> See at (dee3).
>
> On Tue, Aug 20, 2013 at 5:31 PM, Linda Dunbar <[e-mail]> wrote:
> Fang Wei,
>
>
>
> Answers to your comments are inserted below:
>
>
>
>
>
> From: [e-mail] [mailto:[e-mail]]
> Sent: Monday, August 19, 2013 9:57 PM
>
>
> To: Linda Dunbar
> Cc: [e-mail]; [e-mail]
> Subject: RE: [trill] Comments on draft-dunbar-trill-directory-assisted-encap
>
>
>
>
> Hi,Linda
>
> I understand that R1 would keep the encapsulation frame from E1. There would bring two issues:
>
> (1) The remote egress RBridge could not learn the {nickname, MAC} pair by the data frame.
>
> [Linda] The “nickname” is Ingress’ nickname, the MAC-SA is ingress’ too. So the egress RBridge should be able to learn the mapping between the Ingress’ nickname and its MAC. Why do you say “could not”?
>
> (dee3) Right, its asymmetric. So the end station does the ingress but the egress RBridge does the usual egress process, learning the remote MAC address as attached to the ingress nickname in the TRILL Data frame -- but it does not matter that much if the egress RBridge learns anything if you do ingress for traffic going the other way at the end station also.
> (2) The transit RBridges could not pass the RPF check for multi-destination forwarding.
>
> [Linda] Why? The data frame is same as if the frame is encapsulated by the Ingress RBridge.
>
>  (dee3) The only place where there might be an RPF check problem is on receipt at the first edge RBridge so some care may be needed there. Alternatively, you could just not do the ingress at the end station for multi-destination frames but have them ingressed by the edge RBridge.
>
> Thanks,
> Donald
> =============================
> Donald E. Eastlake 3rd   [phone] (cell)
> 155 Beaver Street, Milford, MA 01757 USA
> [e-mail]
> Linda
>
>
>
>
> Linda Dunbar <[e-mail]>
>
> 2013-08-20 10:41
>
>
> 收件人
> "[e-mail]" <[e-mail]>
> 抄送
> "[e-mail]" <[e-mail]>rg>, "[e-mail]" <[e-mail]>
> 主题
> RE: [trill] Comments on draft-dunbar-trill-directory-assisted-encap
>
>
>
>
>
>
>
>
> Fang Wei,
>
> Sorry for the late response due to lack of internet access during my vacation last two weeks. Answers to your questions are inserted below:
>
> From: [e-mail] [mailto:[e-mail]]
> Sent: Friday, August 09, 2013 2:59 AM
> To: Linda Dunbar
> Cc: [e-mail]; [e-mail]
> Subject: Re: [trill] Comments on draft-dunbar-trill-directory-assisted-encap
>
>
> Hi,Linda
>
> When R1 receives a frame, it will check the MAC-SA, and find that it is from E1.  Will R1 re-encapsulate the ingress nickname from "phantom nickname" to N1(R1's nickname)?
> [Linda] R1 will not re-encapsulate the frame because the MAC-DA of the frame is R1.  R1 will terminate the MAC header and look into  the payload, which is a TRILL frame. R1 will put another MAC header with MAC-DA being the egress RBridge’s MAC and MAC-SA being its own MAC.
>
>
>
> In addition, how about the multi-destination forwarding , does the multi-destination forwarding support in this document?
>
> [Linda] For multi destination frames, the destination Nickname should be a root of multicast tree.
>
> Linda
>
>
> Best Regards.
>
> Linda Dunbar <[e-mail]>
> 发件人:  [e-mail]
>
> 2013-08-03 14:39
>
>
>
>
> 收件人
> "[e-mail]" <[e-mail]>cn>, "[e-mail]" <[e-mail]>
> 抄送
> 主题
> Re: [trill] Comments on draft-dunbar-trill-directory-assisted-encap
>
>
>
>
>
>
>
>
>
>
>
>
> Fang Wei,
>
> Sorry for the delayed response. Answers to your comments are inserted below:
>
> From: [e-mail] [mailto:[e-mail]] On Behalf Of [e-mail]
> Sent: Wednesday, July 31, 2013 10:49 AM
> To: [e-mail]
> Subject: [trill] Comments on draft-dunbar-trill-directory-assisted-encap
>
>
> Hi, Linda
>
> I have several comments:
>
> (1) We say that E1 is the "Non-RBridge Nodes", R1 is the local edge RBridge. E1 will encapsulate the TRILL encapsuation with the ingress nickname as "Phantom nickname", and  egress nickname as the nickname of R1(we say it N1). The encapuslation frame will be forwarded to R1. My question is that, how R1 know that the trill encapuslation frame is sent by E1(Non-RBridge node)?
>
> [Linda] There is an Ethernet header in front of the RBridge header. R1 knows that the frame from is from E1 from the MAC-SA.
>
>
> (2) The concept of  'TRILL   Encapsulating node'' or ''Simplified RBridge' is defined in section 3. But it never used in the document. Is it the same meaning with non-RBridge nodes?
>
> [Linda] We call a non-RBridge nodes that can encapsulate TRILL header as “TRILL Encapsulating node”, or “Directory reliant smart end node”. We can finalize the term when going through the WG adoption.
>
> Linda
>
>
>
> Fangwei
>
> Regards_______________________________________________
> trill mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/trill
>
> _______________________________________________
> trill mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/trill
>
> _______________________________________________
> trill mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/trill
>