Hi Robert, Just as Osama pointed out, this draft is to stitch NVO3 tunnel and MPLS VPN Tunnel at DC gateway, it is a transformed option-B defined in RFC4364 which is used for heterogenious network interworking.
As i have known, currently no hardware supports the heterogenious option-B interconnection feature, so we think it's necessary to define the behaviour on DC gateway to provide reference for the industry. Thanks, weiguo ________________________________ From: BESS [[email protected]] on behalf of Robert Raszuk [[email protected]] Sent: Tuesday, April 14, 2015 1:24 To: Osama Zia Cc: [email protected] Subject: Re: [bess] draft-hao-bess-inter-nvo3-vpn Osama, To be very clear ... I am not stating that this draft is inventing VNIDs .. it is NVO3 group strategy I am questioning here :) Also - as I am sure you know - if you look at other communities they also have their own preferences as to tenant demux IDs and encapsulations. All I am trying to point out that the exact functionality this draft describes is already available and deployed without need to invent new gateway behaviors if you happen to use in your data center one of the open source solutions which uses IP forwarding with 4364 as tenant demux. Moreover it works fine today with all 4364 inter-as options A, B, C or even D leave alone other less popular flavors like C+/C- .... Hence IMO one should really think why would we not use it and what other then compliance with NVO3 do we really gain here technically ? Best, r. On Mon, Apr 13, 2015 at 7:12 PM, Osama Zia <[email protected]<mailto:[email protected]>> wrote: Thanks for the clarification Robert. MPLS has been around for quite a while, there are definitely solutions out there like the ones you have mentioned below. However, there are quite a number of deployments out there that do not use MPLS. In fact, if you look at the meeting minutes for NVO3 in IETF-91, you will observe that MPLS was deferred to other WGs and that led to new overlay technologies like GUE, GENEVE apart from existing VXLAN. BTW…NVO3 have just adopted them as working group documents. http://tools.ietf.org/wg/nvo3/minutes?item=minutes-91-nvo3.html I would also like to clarify that VNID is not something we have invented for the purpose of this draft. It is already out there and we are providing a solution that can bridge these two technologies. Regards, Osama From: [email protected]<mailto:[email protected]> [mailto:[email protected]<mailto:[email protected]>] On Behalf Of Robert Raszuk Sent: Monday, April 13, 2015 8:49 AM To: Osama Zia Cc: [email protected]<mailto:[email protected]> Subject: Re: [bess] draft-hao-bess-inter-nvo3-vpn > Are you suggesting to use labels in the CLOS network Nope ... please ... Transport of RFC4364 can be realized with IPv4 or IPv6 just fine. Please see: MPLS in IP or GRE https://tools.ietf.org/html/rfc4023 MPLS in UDP https://tools.ietf.org/html/rfc7510 > instead of any other encap mechanism and use inter-as options in their native > form? Please observe that MPLS in RFC4364 is used just for demux to tenants. Functionally it is identical equivalent to VNID hence inventing new terminology seems in odds with deployed best practices. MPLS for transport on the other hand which some RFC4364 deployments use has no use in IP based CLOS fabric (of course provided that the underlay IP numbering is done with a bit consciousness :) So it is key to understand that transport and tenant demux spaces are completely orthogonal and while some mix them they do not overlap functionally at all. Cheers, R. On Mon, Apr 13, 2015 at 5:32 PM, Osama Zia <[email protected]<mailto:[email protected]>> wrote: Okay, so help me understand your point of view here. Are you suggesting to use labels in the CLOS network instead of any other encap mechanism and use inter-as options in their native form? Regards, Osama From: [email protected]<mailto:[email protected]> [mailto:[email protected]<mailto:[email protected]>] On Behalf Of Robert Raszuk Sent: Monday, April 13, 2015 7:42 AM To: Osama Zia Cc: [email protected]<mailto:[email protected]> Subject: Re: [bess] draft-hao-bess-inter-nvo3-vpn Osama, Yes I have read the draft. It is describing a solution to problem which does not exist in practice. Ref: http://www.opencontrail.org/ The crux of the issue you are solving is here: Each NVE is a BGP speaker. Operator may configure single and unique VNID for each tenant network on all NVEs or configure NVEs to locally allocate VNID for each VRF on the NVEs, the VNID is called VNID-t. If you would not mess up RFC4364 with VNIDs and use vanilla MPLS label as a demux you would not require any additional mapping at the ASBRs hence your option B as well as option C would work seamlessly. All over plain IP CLOS fabric + any type of WAN. That is my point. Best, r. On Mon, Apr 13, 2015 at 3:13 PM, Osama Zia <[email protected]<mailto:[email protected]>> wrote: Robert, I am not sure if you have read the draft. The discussion here is not about using RR. It is about looking at how option B is used in the context of inter-as where both sides are not using MPLS. I would argue that in this solution option-C provides no real benefit. For scaling purposes we are already proposing a centralized solution. Thanks, Osama From: [email protected]<mailto:[email protected]> [mailto:[email protected]<mailto:[email protected]>] On Behalf Of Robert Raszuk Sent: Saturday, April 11, 2015 3:03 AM To: Osama Zia Cc: [email protected]<mailto:[email protected]> Subject: Re: [bess] draft-hao-bess-inter-nvo3-vpn Osama, To the best of my knowledge RRs are pretty base elements in BGP :) If you have 1000 of endpoints within any data center please kindly observe that it is not only ASBR option B information that they need (even if you would establish 1000 IBGP vpnv4/6 sessions to such border router. Such endpoints also need to exchange information between themselves and here again RRs come to the rescue (at least for the time being). So I recommend to look not at one particular function you are pushing, but at entire picture then evaluate the need for RRs likely with RTC or not. Best, r. On Sat, Apr 11, 2015 at 4:15 AM, Osama Zia <[email protected]<mailto:[email protected]>> wrote: I was talking about the base model. You are right that we may solve the scalability problem by introducing RR. However, with the proposed solution we achieve the same thing without using RR. Regards, Osama From: [email protected]<mailto:[email protected]> [mailto:[email protected]<mailto:[email protected]>] On Behalf Of Robert Raszuk Sent: Friday, April 10, 2015 3:42 PM To: Osama Zia Cc: [email protected]<mailto:[email protected]> Subject: Re: [bess] draft-hao-bess-inter-nvo3-vpn Hi Osama, > This would mean the other service provider will need to establish > EBGP sessions possibly in thousands which is not practical. That assertion is not correct. Option C for route distribution scaling uses route reflection in each cluster therefor eliminating the need to establish more then one (or two for redundancy) EBGP sessions between data centers or independent clusters. Cheers, R. On Sat, Apr 11, 2015 at 12:32 AM, Osama Zia <[email protected]<mailto:[email protected]>> wrote: Looking at the minutes, I think there wasn’t enough time left for the draft-hao-bess-inter-nvo3-vpn that a discussion could take place. I will explain the answer to the question that was asked. In option C the ASBR-W will have to establish EBGP will each NVE in the DC. NVE in some cases could be a ToR and in most cases will be a software gateway in the hypervisor. In any case the number of NVEs can run in thousands. This would mean the other service provider will need to establish EBGP sessions possibly in thousands which is not practical. In this scenario Option-B will provide the architecture to address the scaling issue. I would like to request additional comments and call for adoption. Regards, Osama From: BESS [mailto:[email protected]<mailto:[email protected]>] On Behalf Of Susan Hares Sent: Wednesday, April 8, 2015 2:29 PM To: [email protected]<mailto:[email protected]> Subject: [bess] Minutes for IETF 92 BESS meeting The BESS minutes for IETF 92 session are attached (html, pdf). Should you want to make changes to the minutes, you may edit the etherpad location. The minutes have been upload to the etherpad for BESS 92 which is at: http://etherpad.tools.ietf.org:9000/p/notes-ietf-92-bess?useMonospaceFont=true Sue Hares _______________________________________________ BESS mailing list [email protected]<mailto:[email protected]> https://www.ietf.org/mailman/listinfo/bess
_______________________________________________ BESS mailing list [email protected] https://www.ietf.org/mailman/listinfo/bess
