Hi Xu, Today MPLS over GRE support in most implementations I am familiar with has been supported for applications using MPLS labels as demux to VRFs (example L2 or L3 VPNs).
It is not directly obvious that one can claim support for transport MPLS as SID over GRE on existing deployed hardware without requirement to change existing data plane. I mean of course line rate support in line cards rather then punting it to slow path. After all the switching path needs to support following sequence of operation: - Decapsulation of GRE header - POP - Lookup subsequent MPLS label as SID in global table - Encapsulation in GRE + MAC rewrite Clearly software upgrade is needed anyway to support SR, but I think you are making a claim that v6 data plane support will be much harder then the above. I think the answer really depends on the particular hardware used. Regards, R. On Fri, Mar 28, 2014 at 10:30 AM, Xuxiaohu <[email protected]> wrote: > > >> -----邮件原件----- >> 发件人: spring [mailto:[email protected]] 代表 Robert Raszuk >> 发送时间: 2014年3月28日 16:38 >> 收件人: Xuxiaohu >> 抄送: Yakov Rekhter; [email protected]; Alvaro Retana, (aretana) >> 主题: Re: [spring] WG Adoption Call for >> draft-martin-spring-segment-routing-ipv6-use-cases >> >> Hi Xu, >> >> Sure you could. You could do a lot of other things as well. >> >> But what does this buy you if you already have dual stack network which most >> operators already do for years ? > > Even you already have a dual stack or IPv6 network, it doesn't mean you can > run IPv6-SR directly on it w/o any change to the IPv6 data plane. IMHO, the > IPv6-SR approach may be suitable for you to achieve the goal that you have > mentioned when those 4 out of those 1000 nodes have been successfully > upgraded to support IPv6-SR while the approach of combining MPLS-SR and > MPLS-over-GRE could still be considered as a viable way (w/o any change to > the data plane) before the above assumption becomes a reality. > > Best regards, > Xiaohu > >> For clarity I do not consider as sufficient reason fear from one going into >> anaphylactic shock at the mention of IPv6 as a valid one. >> >> Cheers, >> R. >> >> >> On Fri, Mar 28, 2014 at 3:08 AM, Xuxiaohu <[email protected]> wrote: >> > Hi Robert, >> > >> > >> > >> > I think Yakov’s point could be interpreted as “you can use MPLS-SR in >> > conjunction with MPLS-over-GRE to realize the goal that you have >> > mentioned, i.e., only those 4 out of those 1000 nodes need to support >> segment routing. >> > None of the transit nodes needs to be SR aware”. For more details, >> > please see >> > http://tools.ietf.org/html/draft-xu-spring-islands-connection-over-ip >> > and http://www.ietf.org/proceedings/89/slides/slides-89-spring-0.pptx >> > >> > >> > >> > Best regards, >> > >> > Xiaohu >> > >> > >> > >> > 发件人: spring [mailto:[email protected]] 代表 Robert Raszuk >> > 发送时间: 2014年3月27日 23:47 >> > 收件人: Yakov Rekhter >> > 抄送: Alvaro Retana, (aretana); [email protected] >> > 主题: Re: [spring] WG Adoption Call for >> > draft-martin-spring-segment-routing-ipv6-use-cases >> > >> > >> > >> > Yakov, >> > >> > The topic here is traffic controll and traffic steering. In MPLS to >> > the best of my knowledge the only tool to accomplish that is MPLS TE. >> > >> > That in turn requires RSVP-TE and pretty heavy IGP extensions. >> > >> > I am not aware of any deployments of the above MPLS control planes >> > over GRE nor I am aware of using those when only subset of nodes in >> > your domain supports required extensions. >> > >> > If however you are pointing out that to replace LDP one can use GRE I >> > completly agree. >> > >> > Thx, >> > R. >> > >> > On Mar 27, 2014 4:27 PM, "Yakov Rekhter" <[email protected]> wrote: >> > >> > Robert, >> > >> >> Hi Yakov, >> >> >> >> > The above assumes that the already available technology, MPLS, "may >> >> > not be available or deployable for lack of support on network >> >> > elements", yet a brand new technology, a new IPv6 Segment Routing >> >> > header, will be available, deployable, and supported on network >> >> > elements. The authors should elaborate on what makes them think so. >> >> >> >> Let's observe that in order to encapsulate traffic in MPLS in 1000 >> >> node network and steer it north and south 1000 nodes need to be able >> >> to support MPLS (or for that matter RSVP-TE and IGP extensions - >> >> let's leave aside inter-area problems). >> >> >> >> Contrary if the desire is to steer traffic via north and south nodes >> >> from given ingress node to egress node only those 4 out of those 1000 >> >> nodes need to support segment routing. Non of the transit nodes needs >> >> to be SR aware. >> > >> > The ability to carry MPLS over transit nodes that are not MPLS aware >> > is an already solved problem - it is called MPLS-over-GRE (rfc4023). >> > Implemented by multiple vendors. Known to be interoperable. >> > >> > Yakov. >> > >> > _______________________________________________ >> > spring mailing list >> > [email protected] >> > https://www.ietf.org/mailman/listinfo/spring >> > >> > >> > _______________________________________________ >> > spring mailing list >> > [email protected] >> > https://www.ietf.org/mailman/listinfo/spring >> > >> >> _______________________________________________ >> spring mailing list >> [email protected] >> https://www.ietf.org/mailman/listinfo/spring > _______________________________________________ > spring mailing list > [email protected] > https://www.ietf.org/mailman/listinfo/spring _______________________________________________ spring mailing list [email protected] https://www.ietf.org/mailman/listinfo/spring
