Re: [Pce] draft-ietf-pce-pcep-extension-native-ip-14.txt---with HyperLink Correction

[Aijun Wang <[e-mail]>]

Hi, Susan:
Thanks for your review!
Detail responses are inlines below.

Aijun Wang
China Telecom

> On Jul 30, 2021, at 05:50, Susan Hares <[e-mail]> wrote:
> 
> 
> Aijun:
>  
> I apologize for missing PCE session today.  I re-injured a knee just before the session.  Let me provide you suggestions for sections 6.1, 6.2, and 6.3 that may resolve some of the issues.  
>  
> This specification provides one mechanism for BGP auto-configuration.  I’m happy to continue to review this draft and provide suggestions. 
>  
> Cheerily, Susan Hares
>  
>  
> Section 6.1
> =============
> Old text:/
>     The procedures for establishing the BGP session between two peers
>      By using the PCInitiate and PCRpt message pair is show below: /
>  
> New text:/
>     The PCInitiate message can be used to configure the parameters for
>     a BGP peer session using the PCInitiate and PCRpt message pair.
>     This pair of PCE messages is exchanged with a PCE function
>     Attached to each BGP peer which needs to be configured. 
>     After the BGP peer session has been configured
>     via this pair of PCE messages the BGP session establishment process
>     operates in a normal fashion. 
>  
>     All BGP peers are configured for peer to peer communication whether the
>     peers are E-BGP peers or I-BGP peers.   One of the IBGP topologies requires
>     that multiple I-BGPs peers operate in a route-reflector I-BGP peer topology.
>     The example below shows 2 I-BGP route reflector clients interacting
>     with one Route Reflector (RR), but Route Reflector topologies may have
>     up to 100s of clients.  Centralized configuration via PCE provides
>     mechanisms to scale auto-configuration of small and large topologies.  
[WAJ] Will update the draft later accordingly to your suggestions.
>  
> old text:/
>     In the example in Figure 1, if the routers R1 and R7 are within one AS
>     and R3 acts as a router reflector.  PCInitiate message should be sent
>     to route reflector clients R1 (M1) and R7 (M4),a nd the route reflector
>     clients (R3 M2 & M3) respectively.  For inter-AS scenario, such message
>     can be sent directly to ASBR router to build EBGP session./
>  
> New text:/
>     The route reflector topology for a single AS is shown in Figure 1.
>     The BGP routers R1, R3, and R7 are within a single AS.  R1 and R7
>     are BGP router-reflector clients, and R3 is a Route Reflector.
>  
>      The PCInitiate message should be sent all of the BGP routers that
>      need to be configured R1 (M3), R3 (M2 & M3), and R7 (M4). /
[WAJ]  Will update the draft later accordingly to your suggestions.
>  
> ---------------------------
> Section 6.2. 
>  
> My understanding for your email is that the explicit route
> establishment procedures are installing a static route
> which is not to be redistributed via BGP peer.
[WAJ] Yes, the information deployed via EPR object will not redistributed via BGP.
>  
> If this understanding is correct, please clearly state this
> fact and modify the diagrams to indicate a static route
> A suggestion for text for this case is:
>  
> Old text: /
> The detail procedures for the explicit route establishment
> procedures are show below. /
>  
> New text: /
> The explicit route establishment procedures can be used
> to install a route via PCE in the PCC/BGP Peer.
[WAJ] Correct.
> Based on BGP policy, these routes may or may not
> be redistributed by the BGP peers.
[WAJ] Such info will not be redistributed via BGP. The BGP session that established via BPI object will only advertise Prefixes that included in the PPA object.
>  PCE explicit routes
> operate similar to static routes installed
> by network management protocols (netconf/restconf)
> but the  routes are associated with the PCE routing module.   
[WAJ]Correct
>  
> An example of the use case where the of the use case where
> explicit route installed by PCE is redistributed by BGP is
> described in section 6.3. 
[WAJ] No. The information in EPR object is not redistributed by BGP.  Section 6.3 describes how to deploy the prefixes that should be advertised by BGP session that established via BPI object.
The information within PPA has no relation with the information in EPR object.

> An example of the procedure
> where the explicit route installed by PCE is not redistributed
> by BGP is described in this section (see figure 2 below for the
> topology).  
[WAJ]I think this sentence should be omitted.

>  Explicit route installations (like NM static routes)
> must carefully install and uninstall static routes in an specific
> order so that the pathways are established without loops. /
[WAJ] Will update the draft later accordingly.

> ----------------
> Section 6.3
> Old text:/
> The detailed procedures for BGP prefix advertisement are shown
> Below using PCInitiate and PCRpt message pair./
>  
> New Text:/
> The explicit route establishment procedures can be used
> to install a route via PCE in the router that is both
> a PCC and BGP Peer.  Based on BGP policy, these routes   
> may be redistributed by the BGP peers (or may not be distributed).  
>  
> This section provides an example  of the
> PCInitiate and PCRpt messages for the use case when
> routes installed via PCE are redistributed within an IBGP
> cloud with RR and RR clients when a I-BGP policy allows
> PCE explicit routes to be originated within the I-BGP cloud.
> The E-BGP use case and the IBGP use case without
> route reflectors are not shown in this example.  
>  
> The redistribution via BGP of PCE explicit routes
> based on the BGP policy is similar to redistribution
> of static routes installed by network management
> protocols (netconf/restconf) are redistributed by BGP. 
> However, in this use case configured routes are
> associated with the PCE routing module and BGP
> policy must handle the redistribution of these routes. /
[WAJ] The above description is not correct. Wish my explanation can help you to review it again.
Wait for your updated suggestions.

> ==========
>  
>     
>    
>  
>  
>      
>  
> From: Aijun Wang [mailto:[e-mail]] 
> Sent: Wednesday, July 28, 2021 3:42 AM
> To: 'Susan Hares'; [e-mail]
> Cc: 'idr-chairs'; 'pce-chairs'
> Subject: RE: [Pce] draft-ietf-pce-pcep-extension-native-ip-14.txt---with HyperLink Correction
>  
> Hi, Susan, Mike and All other experts:
>  
> I have just upload the updated version of this draft, would you like to review it again to refine it?
> And you can also refer the presentation on this IETF meeting for additional update introduction(https://datatracker.ietf.org/meeting/111/materials/slides-111-pce-sessa-21-native-ip-00.pdf).
> Thanks in advance.
>  
>  
> Best Regards
>  
> Aijun Wang
> China Telecom
>  
> From: Susan Hares <[e-mail]> 
> Sent: Wednesday, July 28, 2021 5:58 AM
> To: 'Aijun Wang' <[e-mail]>; [e-mail]
> Cc: 'idr-chairs' <[e-mail]>; 'pce-chairs' <[e-mail]>
> Subject: RE: [Pce] draft-ietf-pce-pcep-extension-native-ip-14.txt---with HyperLink Correction
>  
> Aijun:
>  
> Thank you for your quick response.  
>  
> I’ve answered questions below.  Would you let me know when you update your draft?  I’ll re-read the sections that changed and make additional suggestions.  If you think it would be useful to add 2 (or more) next hops per EPR, please let me know.
> [WAJ] Adding 2 (or more) next hops per EPR is one viable solution to achieve the ECMP effects, but it seems that the current format is more flexible and extensible? For example, if we want to add the UCMP features, as Mike proposed, we can add some information within the “Optional TLV” field of the EPR to indicate what percentage traffic should be allocated to each nexthop.
>  
> I’m glad to keep reviewing your changes.
>  
> Cheers, Sue
>  
> From: Aijun Wang [mailto:[e-mail]] 
> Sent: Tuesday, July 27, 2021 3:19 AM
> To: 'Susan Hares'; [e-mail]
> Cc: 'idr-chairs'; 'pce-chairs'
> Subject: RE: [Pce] draft-ietf-pce-pcep-extension-native-ip-14.txt---with HyperLink Correction
>  
> Hi, Susan and all:
>  
> Update one hyperlink on the contents, please refer to this mail for comments responses.
>  
> From: [e-mail] <[e-mail]> On Behalf Of Aijun Wang
> Sent: Tuesday, July 27, 2021 12:04 PM
> To: 'Susan Hares' <[e-mail]>; [e-mail]
> Cc: 'idr-chairs' <[e-mail]>; 'pce-chairs' <[e-mail]>
> Subject: Re: [Pce] draft-ietf-pce-pcep-extension-native-ip-14.txt
>  
> Hi, Susan:
>  
> Thanks for your reviews, let me first address your current questions and wait for the further discussions on the overall solution.
>  
> Best Regards
>  
> Aijun Wang
> China Telecom
>  
> From: [e-mail] <[e-mail]> On Behalf Of Susan Hares
> Sent: Tuesday, July 27, 2021 7:19 AM
> To: [e-mail]
> Cc: 'idr-chairs' <[e-mail]>; 'pce-chairs' <[e-mail]>
> Subject: [Pce] draft-ietf-pce-pcep-extension-native-ip-14.txt
>  
> Greetings:
>  
> Thank you for your work on draft-ietf-pce-pcep-extension-native-ip-14.txt.   
> This comment should be consider feedback from me as a WG member of IDR and PCE.  I have posted this information also on the  
>  
> This draft takes a step toward auto-configuration of BGP peers.  IDR has created a set of requirements for BGP auto-configuration for Data Centers at:
> https://datatracker.ietf.org/doc/draft-ietf-idr-bgp-autoconf-considerations/
> [WAJ] Yes, I have also reviewed this document and attended some discussions on it. The autoconf document tries to build the BGP from scratch without 3rd party(for example, PCE) assistance. It considers mainly the direct connected BGP peer setup process automatically and does not involve the prefix advertisement and explicit route setup.
> I think the aim of these two drafts is different but we can refer to some designed considerations from it.
>  
> [Sue] You have understood that the bgp auto-configuration works on peer set-up.
>           I simply wanted you to know that your PCE work linked that that work in BGP.
>  
> I’ve put a copy of these comments at:
> https://trac.ietf.org/trac/idr/wiki/pce-pcep-extension-native-ip%20Hares%20comments
>  
>  
> Cheers, Sue
>  
> Comments on draft-ietf-pce-pcep-extension-native-ip-14
> =================================================
> Overview of errors
> 1) section 6 description of BGP routers needs clarification
> (sections 6.1, 6.2, and 6.3) for RR and RR Clients
> [WAJ] Please see the replies inline below.
>  
> 2) BGP Session Establish Procedures – are these restrict to RR and RR Clients?  
> [WAJ] Yes. The BGP session is established between RR and its clients in large network. It can also be established between two nodes directly.(as described in https://datatracker.ietf.org/doc/html/rfc8821#section-3)
>  
> The point –to-point (P2P) Peering was clear in RFC8821.
> [Sue] It would be useful to revise the draft to clearly define this is a single AS,
> RR location, and RR client locations.
> [WAJ] Has added some text as below. Actually, the procedures described in this draft is not limited to one AS, it can be applied to inter-AS scenario simultaneously.
> “In the example in Figure 1, if the router R1 and R7 are within one AS and R3 acts as the route reflector, PCInitiate message should be sent to route reflector clients R1(M1) and R7(M4), and the route reflector R3(M2 & M3) respectively. For inter-AS scenario, such message can be sent directly to the ASBR router to build EBGP session.”
>  
> 3) Explicit routes [section 6.2] – Is ECMP support as well as 1 prefix/1 next Hop?  
> [WAJ] Yes, ECMP is supported. PCE needs to send two EPR(Explicit Peer Route) objects, with the same “Destination Adress” and “Route Priority”, but different “Next Hop Address”
> [Sue] – it is possible to be more effective in the EPR by including 2 (or more)
> Next Hops.  Have you consider this choice in PCE?
> [WAJ] As I answered in the begin of this email, it seems to encode one single hop per EPR is more clear, extensible and flexible?
>  
> 4) IPv4/IPv6 restrictions [section 6.3] – are you restricting the peer session or the AFI/SAFI supported by the Peer session?
> [WAJ] AFI/SAFI supported by the peer session.
> [Sue] Again – it would be helpful to clearly state AFI/SAFI.
> [WAJ] Has updated the draft accordingly, as “The allowed AFI/SAFI for the IPv4 BGP session should be 1/1(IPv4 prefix) and the allowed AFI/SAFI for the IPv6 BGP session should be 2/1(IPv6 prefix). If mismatch occur, an error(Error-type=TBD6, Error-value=TBD18, BPI/PPR address family mismatch) should be reported via PCErr message.”
>  
> 5) Sections 7, 9, and 10 – may need to change based on your answers to questions 1-4?
>  
> Detailed questions
> ---------------------
> 1. Section 6 – sub sections 6.1, 6.2 and 6.3.
>  
> Problem:
> The text that describe the BGP peers and the diagram needs clarification on the BGP peering between BGP peers:  R1, R7, and R3. If R1 and R7 are Route Reflector clients (RR clients) are attached to the R3 then it is important to indicate this point. 
> [WAJ] Yes, R1 and R7 is RR clients.
> [Sue] Please make this fact clear in your next revision.
> [WAJ] Done
>  
> If you are using classic route reflection, then R1, R3 and R7 would need to be in the same Autonomous system.
> [WAJ] Yes, certainly.
> [Sue] Please make that clear in your document.
> [WAJ] Done
>  
> The RR (R3) determines what routes are sent to the RR clients. 
>  
> This problem impacts the text in section 6.1, 6.2, and 6.3
>  
> 2.) Text change for Section 6.1 – if R1 and R7 are RR clients.
>  
> Here’s a change if R1 and R7 are Router Reflector Clients. 
>  
> Current text: /
>    The PCInitiate message should be sent to PCC which acts as BGP router
>   and route reflector(RR).  In the example in Figure 1, it should be
>    sent to R1(M1), R3(M2 & M3) and R7(M4), when R3 acts as RR./
>  
> Improved text: /
>    The PCInitiate message should be sent to PCC which acts as
>     BGP router reflector or a route reflector client. In the
>    example in Figure 1, it should be
>    sent to the route reflector clients R1(M1) and R7 (M4), and
>   the route reflector R3 (M1 or M4).    /
> [WAJ] Has updated the draft with the following contents(almost same as your suggestions):
> The PCInitiate message should be sent to PCC which acts as BGP router and/or route reflector (RR). In the example in Figure 1, it should be sent to route reflector clients R1(M1) and R7(M4), and the route reflector R3(M2 & M3) when R3 acts as RR. PCInitiate message creates an auto-configuration function for these BGP peers providing the indicated Peer AS and the Local/Peer IP Address.
> [Sue] Thank you.  Did you submit your updated draft?
> [WAJ] Done
>  
> 3) Section 6.1 – BGP Session Establishment Procedure
> Question: Does the PCEInitiate (message and report) require the RR and RR client structure?
> [WAJ] No. not necessary.  The BGP session setup procedures is same between RR and its clients.
> [Sue]: If you use the PCEInitiate message with another structure, it would be useful
> To include an example in your document of multiple peers.
> [WAJ] The BPI(BGP Peer Info) object is sent only to the two BGP peers, will not be sent to multiple peers (>2) simultaneously.
>           The PPA(Peer Prefix Advertisement) object is sent only to the prefixes source node(for example, R1 or R7 in Figure 4). The RR needs not receive such instructions. RR will receive the advertised prefix and forward it to its other clients, based on the normal RR behavior.
> Then, there is no differences for the BPI and PPA messages to RR, or its clients.
>  
> If so, the PCInitiate should have a parameter indicate what type of BGP peer (RR or RR client) each receiving BGP peer should be.   
>  
> 4) Section 6.2 – Explicit Route Establishment Procedure
>  
> Problem: It is unclear what the impact to the routing system of the setting of explicit route. 
>  
> Basic Details: (1 Route with 1 Next Hop)
> If R1 and R7 are RR clients and the Explicit route operates as static route installed by the PCIntiate, then BGP peer will reflected these static routes  R3.
> [WAJ] No, the explicit route is only installed on the aimed PCC nodes and such information will not be advertised via the BGP session between RR and its clients.
> [R1 (explicit route [static route]) à R3]
> [R1 (explicit route [static route]) à R7] [note my error]  
>  
>                 [Sue] If a route is installed in the RR clients, why is it not updated to the BGP
>                Peer web (R1-RR-R7). I thought the purpose of adding it to the RR client.
>                 If it is not the purpose, why is this included in the BGP section?
> [WAJ] EPR object is different with BPI and PPA objects.  EPR object is just used to indicate/influence how to get the BGP nexthop.  Such information does not spread via the BGP mechanism, it is deployed by the PCEP protocol directly.
>                               
>  
> Setting or clearing the Explicit route seem to map to a setting/clear a static route on the node.  If this is true, then this section needs to be rework to clear describe the process.
> [WAJ] Yes, it is similar with the static route on the node. The purpose of these explicit routes are to influence the final recursive forwarding path for prefixes advertised by BGP peer.
>  
> [Sue] Then is the purpose to send it from 1 RR client (for example R1), and
>  Have it redistribute to R3 (RR) and R7 (RR client).
> [WAJ] No, it is not redistribute to R3. EPR object is deployed directly via the PCEP protocol. The spread of such information is different from PPA, which is via the BGP control plane.
>  
> Your setting the route on the pathway hop by hop is similar to netconf/restconf setting routes in a pathway.   
> [WAJ] Yes, it is similar.
> [Sue] Thank you.  At least I got one detail correct [smile].
> [WAJ] It’s my responsibility to describe the mechanism more clearly.
>  
> ECMP Details: (1 Route with multiple Next Hop)
> If the explicit route is a ECMP route with multiple next hop paths, the next hop for a route installed in could be R5 or R2.
> [WAJ] Yes. For ECMP routes, the PCE needs to send two EPR objects to PCC(in Figure2, on R1), with the “Destination Address” are both set to R7, but the “Nexthop Address” is set differently(for example, R2, R5 as you mentioned.).  The “Route Priority” field in EPR object should also be the same.
> [Sue] Thank  you for the explanation.  Do you think it is useful to allow 2 Next-Hops
>  In the same EPR object.
> [WAJ] Currently, I think one next-hop per EPR maybe more clear, extensible and flexible. If there is any other strong advantages, we can modify it later.
>  
> If ECMP is allowed, then you need to decide if:
> a) adding this route allows the route to be installed if only some of the next hops are valid (for example R5 is valid, but R2 is not)
> b) delete routing allows the route to be deleted if both next hops were not installed.
> [WAJ]  Adding the description “The PCC should verify that the next hop address is reachable.”  before the sentence “Upon the error occurs, the PCC SHOULD send the corresponding error via PCErr message, with an error information… …。
> [Sue]:  This is a wonderful addition.
>  
> 5) Section 6.3
>  
> Problem:  You do not clear indicate the status of BGP peer routers.
>  
> If R1 and R7 are BGP route reflector clients, then R1 and R7 will send the route to R3 which will reflect the route to other RR clients (if policy allows).
> [WAJ] The propagation of BGP prefixes is the same as the traditional BGP procedures. The “Peer Address” in PPA objects indicates which peer the prefixes will be sent to.
> [Sue] If you have trio of 2 RR clients and RR, then simply indicate you follow RR-RRclient rules.
> [WAJ] The PPA(Peer Prefixes Advertisement) object need only be sent to the source node, then such prefixes information will be flooded based on the regular BGP RR mechanism. We need not send PPA object to all of the RR’s clients.
>  
> 6.) Section 6.3
> Problem:  It is unclear why there is a restriction for IPv4 prefix to be sent only via a IPv4 BGP section, and the IPv6 prefix only via a IPV6 section.
>  
> Details: I think the author is trying to describe the peers support for a particular set of AFI/SAFIS for NLRI sent rather than the peering.  However, it is unclear.
> [WAJ] What we want to express is that the IPv4 BGP Peer session will advertise/receive only IPv4 prefixes(AFI/SAFI is 1/1 ), and IPv6 BGP Peer session will advertise/receive IPv6 prefixes(AFI/SAFI is 2/1).
>                  [Sue]: Why did you make this restriction?  Is it from PCE requirements?
> [WAJ] Just want to simplify the management of network and less error-prune.
>  
> 7.) Sections 7.2 and 7.3
> All of these issues on the intent of the protocol need to be answered before I can provide additional feedback on the PCEP objects.  
>  
> The initial shape of the PCE discussion are reasonable, but working through the details requires clarity in sections 6.1 to 6.3.  For example, support for ECMP in the explicit routes may cause sections 7.3 and 7.4 to be rewritten.
>  
> 8.) Section 9 –
> The error handling must consider the RR to RR client distribution of routes.
> [WAJ] The route distribution process between the RR and RR clients is unchanged.
>  
> Also, if one PCE overwrites another multiple route are sent from a RR client to the RR.  The policy in the RR must be set-up to handle errors.
> [WAJ] The information from EPR object is not advertised by the RR client back to the RR.
> [Sue] What prevents the EPR object from being sent to the RR via BGP.
> Is the EPR object not installed in the BGP RIBs to be sent?
> [WAJ] EPR object is deployed along the optional path via PCEP protocol, it will be installed in the RIB of node, but not the BGP RIB of the node.
> I think I am missing something simple here.
> [WAJ] The EPR object is described between BPI and PPA object section, maybe this lead to your impression that it is also related to the BGP. If necessary, we can adjust the description sequences of these objects later.
>  
> This section needs a bit of rethinking.
>  
> 8.) Section 10 - BGP Considerations -  
> The content of the BGP consideration sections seems reasonable, but it should be reviewed again after all the remainder of the document has been clarified.
> [WAJ] Wish to receive your more through considerations for the current solution.           
>    
> [Sue] Thank you for the discussion.  Could you provide updates