Re: [spring] 答复: Re: WG adoption requested for draft-filsfils-spring-sr-recursing-info

> On Aug 25, 2016, at 4:41 AM, peng.shaofu@zte.com.cn wrote:
> 
> Stefano, 
> 
> see inline with [Deccan] 
> 
> Thanks 
> Deccan 
> 
> 
> 
> "Stefano Previdi (sprevidi)" <sprevidi@cisco.com>
> 2016-08-23 23:22
> 
> 收件人
> "peng.shaofu@zte.com.cn" <peng.shaofu@zte.com.cn>,
> 抄送
> SPRING WG <spring@ietf.org>
> 主题
> Re: [spring] WG adoption requested for draft-filsfils-spring-sr-recursing-info
> 
> 
> 
> 
> 
> Hi Deccan,
> 
> 
> > On Aug 23, 2016, at 4:39 AM, peng.shaofu@zte.com.cn wrote:
> > 
> > Hi Stefano 
> > 
> > Sure, SRRI provided in this document can explicitly indicate a recursive operation (or relationship). 
> > But it maybe a default behavior to do recursive operation when an SR-NODE received remote prefix-sid with L/V flag set according to the documents already existed. For example, 
> > prefix reachability advertisement received: 
> >     prefix (1.0.0.99/32) 
> >         prefix-sid (30004), L/V flag set,  //ISIS-SR 
> >         "IPv4 Source Router ID" is 1.0.0.4 //rfc7794 
> > Then, prefix 1.0.0.9/32 can do recursion to prefix 1.0.0.4/32 by default.
> 
> 
> there are other cases where V/L are used so I wouldn’t 
> bind these flags to the recursive behavior.
> 
> [Deccan] Yes, it is entirely possible to generate a segment list only including segments with local meaning, such as adjacency segment list, service segment list, etc., without any node segment. So it is not appropriate to do recursive operation for local meaning segments by default. There maybe other cases that I don't know.
> 
> > If "default behavior" is not accepted, we can define a new RECURSIVE flag in Prefix-SID Sub-TLV. 
> 
> 
> if I understand your proposal, you could:
> . attach the SourceRouterID (RFC7794) to the prefix
> . attach a prefix-sid with:
>   . recursive flag set, which means the sid is to be 
>      taken from the sourceRouterID
>   . optionally set the V/L flag and also a local label
>   . if no V/L flag are set, the value of the sid/label 
>      field in the prefix-sid must be 0 on transmit and 
>      ignored on receive.
> 
> [Deccan] The above last point could be modified: 
> In despite V/L flag set or not, the value of the sid/label field in the prefix-sid must always be a valid value. That is, for recursive use-case, it is its own sid; for sharing use-case, it is the sharing sid. However, if the received nodes don't know RECURSIVE flag, for sharing use-case, sid conflicting will process; for recursive use-case, no recursion to be done. 

so, it’s broken.

> I see at least 2 problems with that:
> 1. the value of the prefix-sid you’re going to advertise 
>   will be misinterpreted by non-srri-capable nodes 
>   (i.e.: nodes know knowing the recursive flag). I.e: The 
>   value of the prefix sid would be either an explicit 
>   null label either a 0-value index.
> 2. RFC7794 states that: the Router ID advertised is 
>   always the Router ID of the IS-IS instance that 
>   originated the advertisement.
> 
>   This reduces the ability to use other addresses of the 
>   node for recursion.
> 
> Bottom line, having a separate repository for the recursing 
> information is the safest and cleanest option which allows 
> better flexibility (i.e.: allowing to recurse to a 
> non-router-ID address) and which ensure backward 
> compatibility (i.e.: non-srri-nodes will simply ignore the 
> whole srri info and consider the prefix as it had no sid at 
> all).
> 
> [Deccan] Yes, the alternate method only allows to recurse to a router-id address. But, don't you think router-id address is enough for any segment list?

absolutely not. Yu need to be able to recurse multiple prefixes to different addresses (all belonging to the same node).

> Especially for an SR-TE path dynamic computed by CSPF, I think a non-router-id address will have no chance to represent a node-segment. However, the alternate method seems to be not much clean than SRRI method, for the possible sid conflicting introduced in sharing use-case, from this point, the SRRI method is good. 

CSPF is not the main use case here.

> > BTW, all we discussed is SID recursive but not sharing,
> 
> 
> of course it is sharing. As defined in the draft, the 
> local label attached to the srri info it’s an optional 
> optimization. Without the local label, you will share the 
> same sid among multiple prefixes. 
> 
> 
> > even the first case in this draft is actually not SID sharing, otherwise it will be cared by draft-ietf-spring-conflict-resolution. 
> 
> 
> No, it is not a conflict. Having a dedicated srri 
> repository makes it clear.
> 
> [Deccan] Ok, I have not got this point before. However, if router-id address is enough,

router-ID is not enough.

> sid sharing seems to be not so much significant, do you think so? 

I think the srri method is the cleanest and safest option to provide both:
. recursion of a prefix to a different sid
. sharing the same sid among different prefixes
without incurring well known backward-compatibility and conflicting issues.

srri is a generic solution applicable to a variety of use cases.

Thanks.
s.

> 
> s.
> 
> 
> > Thanks 
> > Deccan 
> > 
> > 
> > 
> > 
> > 
> > "Stefano Previdi (sprevidi)" <sprevidi@cisco.com>
> > 2016-08-22 21:38
> > 
> > 收件人
> > "peng.shaofu@zte.com.cn" <peng.shaofu@zte.com.cn>,
> > 抄送
> > SPRING WG <spring@ietf.org>
> > 主题
> > Re: [spring] WG adoption requested for draft-filsfils-spring-sr-recursing-info
> > 
> > 
> > 
> > 
> > 
> > 
> > > On Aug 9, 2016, at 5:55 AM, peng.shaofu@zte.com.cn wrote:
> > > 
> > > Other documents have already addressed this issue,
> > 
> > 
> > I don’t think so. Can you point to these documents ?
> > 
> > 
> > > for example, set L-flag of Prefix-SID Sub-TLV in draft-ietf-isis-segment-routing-extensions-05 and contain IPv4 Source Router ID in rfc7794. 
> > 
> > 
> > the L flag has the solely purpose of indicating the sid contains a local value. Typically it goes with the V flag that indicates a value (i.e.: local label).
> > 
> > Nothing is mentioned regarding sharing the same sid among different services.
> > 
> > s.
> > 
> > 
> > 
> > > 
> > > 
> > > Thanks, 
> > > 
> > > Deccan 
> > > 
> > > 
> > > 
> > > 
> > > 
> > > [spring] WG adoption requested for draft-filsfils-spring-sr-recursing-info 
> > > "John G. Scudder" <jgs@juniper.net> Sun, 24 July 2016 12:54 UTCShow header
> > > Dear WG,
> > > 
> > > As we discussed at our meeting, working group adoption has been requested for draft-filsfils-spring-sr-recursing-info. Please reply to the list with your comments, including although not limited to whether or not you support adoption. Non-authors are especially encouraged to comment.
> > > 
> > > We will end the call on August 31, 2015. 
> > > 
> > > Authors, please indicate whether you are aware of any relevant IPR and if so, whether it has been disclosed.
> > > 
> > > Thanks,
> > > 
> > > --Bruno and John
> > > 
> > > 
> > > _______________________________________________
> > > spring mailing list
> > > spring@ietf.org
> > > https://www.ietf.org/mailman/listinfo/spring
> > > 
> > 
> > 
> 
> 
> _______________________________________________
> spring mailing list
> spring@ietf.org
> https://www.ietf.org/mailman/listinfo/spring