Ali, I don’t care about whether or not your draft has a reference to MPLS over UDP. However, since you’ve made a U-turn on the load-balancing capability of most existing core switches, I’m just wondering whether the ship you mentioned below just sailed last month:)
Best regards, Xiaohu 发件人: Ali Sajassi (sajassi) [mailto:[email protected]] 发送时间: 2012年12月12日 13:47 收件人: Xuxiaohu; Lucy yong; Aldrin Isaac; Melinda Shore 抄送: [email protected] 主题: Re: [nvo3] [mpls] MPLS-RT review of draft-xu-mpls-in-udp-03 Xiaohu, First, my draft got merged with draft-drake-nvo3-evpn-control-plane and the resultant draft is now draft-sajassi-drake-l2vpn-evpn-overlay-00.txt. And the new draft has no reference to MPLS over UDP. Second, even in my old draft that you are referencing, if you look at the section that talks about MPLS over UDP, it says: [MPLSoUDP] discusses using a UDP header instead of the GRE header to transport MPLS client layer over an IP PSN tunnel. The main advantage for doing so is for better load-balancing capabilities over existing IP networks, where some core routers can perform ECMP based on the UDP header but not based on the GRE Key field. However, the routers that are capable of supporting [NVGRE<http://tools.ietf.org/html/draft-sajassi-nvo3-evpn-overlay-01#ref-NVGRE>] encapsulation, can also perform load-balancing based on the GRE key which accommodates a 32- bit entropy value; whereas, UDP encapsulation accommodates a 16-bit entropy value. And this sums up why I don't see a value in MPLS over UDP encap (as I have been explaining on this thread). -Ali From: Xuxiaohu <[email protected]<mailto:[email protected]>> Date: Tuesday, December 11, 2012 12:05 AM To: Cisco Employee <[email protected]<mailto:[email protected]>>, Lucy yong <[email protected]<mailto:[email protected]>>, Aldrin Isaac <[email protected]<mailto:[email protected]>>, Melinda Shore <[email protected]<mailto:[email protected]>> Cc: "[email protected]<mailto:[email protected]>" <[email protected]<mailto:[email protected]>> Subject: re: [nvo3] [mpls] MPLS-RT review of draft-xu-mpls-in-udp-03 [Lucy] if MPLSoUDP and NVGREoUDP, underlying network only needs to handle UDP based load balancing, which presents a simpler solution for DC infrastructure. DC do not have to change today’s DC device at all. This is the architecture metric I see. How many DC routers support GRE based load balancing today? Lucy That ship has already sailed and the capability to do ECMP based on GRE key has already been incorporated into ASICs which enables ECMP based on standard-based MPLSoGRE (RFC2890 and RFC2784). Anyway, let's see what DC operators have to say about this. [Xiaohu] Ali, The following text is quoted from your draft (http://tools.ietf.org/html/draft-sajassi-nvo3-evpn-overlay-01) which was just submitted no more than two months ago, “ As noted previously, there are existing core switches that do not support ECMP by hashing the GRE key; however, vast majority of existing core switches support ECMP by hashing UDP header; therefore, VXLAN encapsulation can provide better ECMP functions for these existing switches”. Are you going to correct the above statement in a manner of U-turn in the next revision? Xiaohu [/Xiaohu] -Ali
_______________________________________________ nvo3 mailing list [email protected] https://www.ietf.org/mailman/listinfo/nvo3
