Hi Greg, If C and A-B are statically programmed, VRRP can be useful to indicate to node C which node (A or B) is active for a given service. But, it does not help in scenarios where A-B are doing dynamic routing with node C and the IPs on which the services are being run themselves are dynamically advertised or withdrawn based on need. There are existing mechanism in dynamic routing to program path towards active nodes and BFD can be added to protect a static/dynamic session. Bottomline, we are not defining how to notify C in this draft.
Failure of BFD session between A&B may happen if there is a real node failure or in case of a split brain as you are suggesting. Mechanisms to avoid/fix split brain scenarios for a set of nodes providing redundancy are beyond the scope of this draft. But when a split brain resolves, the methods described in the draft will ensure that a given service is active only on one node. Thanks, --Ankur From: Greg Mirsky <gregimir...@gmail.com> Date: Tuesday, November 28, 2017 at 2:59 PM To: Sami Boutros <sbout...@vmware.com> Cc: Ashesh Mishra <mishra.ash...@outlook.com>, Ankur Dubey <adu...@vmware.com>, "rtg-bfd@ietf.org" <rtg-bfd@ietf.org>, Reshad Rahman <rrah...@cisco.com> Subject: Re: Service Redundancy using BFD Hi Sami, you've indicated that it is that one of the set of network functions (NF), A and B in the figure below, that provides L2/L3 services to NF C. My question was how C addresses the designated forwarder (DF) of the A-B set. If it uses virtual address that associated with the function of the DF, then NF C doesn't need to know the identity of the DF (similar to VRRP, isn't it). If NF C needs to know the identity of the DF, then it must use some means to monitor liveliness of A and B. And I have to point to couple BFD related assumptions in the draft: * failure of BFD session between A and B cannot be interpreted as failure of A or B by respective BFD peer but only as loss of continuity between the forwarding engines. Assumption that the failure is not of link but of a node may lead to duplicate DFs; * using multi-hop BFD to detect node failure may produce false negative if failure detection is more aggressive than network convergence, e.g. network convergence is guaranteed within 100 ms while BFD interval is 10 ms. Regards, Greg On Tue, Nov 28, 2017 at 2:39 PM, Sami Boutros <sbout...@vmware.com<mailto:sbout...@vmware.com>> wrote: Hi Greg, A can detect failures to the link to C using any mechanisms not only BFD. The picture below is for illustration, A and B themselves can be providing services (L4 to L7), this could include Firewall, NAT, LoadBalancer etc.. Thanks, Sami From: Greg Mirsky <gregimir...@gmail.com<mailto:gregimir...@gmail.com>> Date: Tuesday, November 28, 2017 at 2:20 PM To: Sami Boutros <sbout...@vmware.com<mailto:sbout...@vmware.com>> Cc: Ashesh Mishra <mishra.ash...@outlook.com<mailto:mishra.ash...@outlook.com>>, Ankur Dubey <adu...@vmware.com<mailto:adu...@vmware.com>>, "rtg-bfd@ietf.org<mailto:rtg-bfd@ietf.org>" <rtg-bfd@ietf.org<mailto:rtg-bfd@ietf.org>>, Reshad Rahman <rrah...@cisco.com<mailto:rrah...@cisco.com>> Subject: Re: Service Redundancy using BFD Hi Sami, would C have BFD sessions to A and B respectively or it use anycast address? The more I look at the use case, the more I think of VRRP ;) Regards, Greg On Tue, Nov 28, 2017 at 2:15 PM, Sami Boutros <sbout...@vmware.com<mailto:sbout...@vmware.com>> wrote: Hi Ashesh, The topology is more like the following: A <—\ | \ BFD C | / B<—/ A and B are nodes providing L2 and L3 services for C, with A/S redundancy. A can be active and B standby, if A goes down then B start providing the services. Thanks, Sami
