Hi Aldrin, I can hardly understand your question. Would you please explain it in more details?
Best regards, Xiaohu 发件人: Aldrin Isaac [mailto:[email protected]] 发送时间: 2012年9月22日 22:44 收件人: Xuxiaohu 抄送: Jon Hudson; Lucy yong; Kireeti Kompella; [email protected] 主题: Re: [nvo3] Is the cost of supporting active-active multi-homing on a per L2VPN instance basis sustainable? Xiaohu, Assume a Virtual LAN is formed by the interconnection of an L2 VN in an overlay domain with other LAN technology (dot1q, TRILL, PBB, etc) and/or other L2 VN in other overlay domain. In such a case would you mandate that the interconnection points must be active/standby? What then if the VN "bridges" with the VN in the other Overlay domain via multiple sites/DC? Would you pick one site to be the active interconnection point in order to avoid data plane loops? What then of optimizing ingress/egress? What of pooling? Best regards. -- aldrin On Saturday, September 22, 2012, Xuxiaohu wrote: Yes, if the active-active multi-homing on a per L2VPN instance basis was desired in some cases, vendors could offer their proprietary schemes like Cisco’s VSS to support that capability. In this way, there is no need to introduce the complexity into the standard, especially on the data plane perspective, and therefore other NVEs (i.e., PE routers) which are not within any active-active redundancy group don’t need to afford that complexity anymore. As for active-passive multi-homing, I fully agree with you that the active/passive multi-homing on a per physical interface or trunk interface basis is not welcomed. Hence we could achieve load-balancing among the physical interfaces or trunk interfaces which are connected to different NVEs of the same RG (redundancy group) on the basis of VLANs. For example, a dual-homed device (DHD) is multi-homed to NVE1 and NVE2 and it is configured with two VLANs (say VLAN x and y) on its physical NICs in a bond, NVE1’s corresponding interface is active for VLAN x while NVE2’s corresponding interface is active for VLAN y… Best regards, Xiaohu - 发件人: Jon Hudson [mailto:[email protected]<javascript:_e(%7b%7d,%20'cvml',%20'[email protected]');>] 发送时间: 2012年9月22日 17:16 收件人: Xuxiaohu 抄送: Aldrin Isaac; Lucy yong; Kireeti Kompella; [email protected]<javascript:_e(%7b%7d,%20'cvml',%20'[email protected]');> 主题: Re: [nvo3] Is the cost of supporting active-active multi-homing on a per L2VPN instance basis sustainable? If we don't do it as part of the standard, vendors will just implement proprietary versions. active/passive anything is not welcomed in the datacenter anymore On Sat, Sep 22, 2012 at 2:02 AM, Xuxiaohu <[email protected]> wrote: Hi all, I strongly believe this is an very important question which needs more attentions and discussions. Hence I redefine the subject of the original email so as to avoid this email from being missed. Best regards, Xiaohu > -----邮件原件----- > 发件人: [email protected] [mailto:[email protected]] 代表 > Xuxiaohu > 发送时间: 2012年9月21日 10:00 > 收件人: Aldrin Isaac; Lucy yong > 抄送: Kireeti Kompella; [email protected] > 主题: [nvo3] 答复: draft-drake-nvo3-evpn-control-plane > > > > > -----邮件原件----- > > 发件人: [email protected] [mailto:[email protected]] 代表 > Aldrin > > Isaac > > 发送时间: 2012年9月19日 5:24 > > 收件人: Lucy yong > > 抄送: Kireeti Kompella; [email protected] > > 主题: Re: [nvo3] draft-drake-nvo3-evpn-control-plane > > > > The imported E-VPN route will determine what the next hop entry in the > > EVI will look like -- whether it will have encapsulation A or > > encapsulation B. That is determined by the sender of the E-VPN route. > > This is like having a PPP interface and an Ethernet interface > > connected to the same VRF. > > > > Ideally the other encapsulations would have included support for > > multi-homing by including an additional field for a split-horizon ID > > for use by control-planes and NVE that support multi-homing. Maybe > > it's not too late to add an SH field to these encapsulations since it > > seems that there is some unused bits in nvGRE (maybe not enough) and > > in VXLAN -- just a thought. > > If the SH field is just intended for supporting active-active multi-homing on > a per > VPN instance basis, I suggest it'd better for us to seriously evaluate whether > the cost for realizing that goal is worthwhile since that will make the > solution > itself, especially the data plane process, much more complex. In the > multi-tenant cloud data center environment, isn't it good enough in practice > to > support active-standby multi-homing on a per VPN instance basis while > supporting active-active multi-homing on a per physical interface basis? For > instance, a physical server containing two VMs (say VM1 and VM2) which > belong to different VPN instances respectively is dual-homed to two NVEs (say > NVE1 and NVE2), NVE1 is the active NVE for VM1 and the standby NVE for VM2 > while NVE2 is the active NVE for VM2 and the standby NVE for VM1. > > Best regards, > Xiaohu > > > On Tue, Sep 18, 2012 at 4:18 PM, Lucy yong <[email protected]> wrote: > > > Hi Kreeti, > > > > > > Regarding interworking
_______________________________________________ nvo3 mailing list [email protected] https://www.ietf.org/mailman/listinfo/nvo3
