Bfd

EdTheLadEdTheLad Member Posts: 2,111 ■■■■□□□□□□
Just wondering about BFD configuration which can vary per platform.
I've got 2 CRSs connected via a bundle-eth interface. I configure bfd on the bundle, on the CRS it gives the option to specify the destination ip address via interface config mode .
The question is, do i need to enable bfd on my IGP, lets say ospf? I presume once bfd detects a failure it will quickly notify ospf so there wouldn't be a need. When using physical interfaces i don't get the option to specify a bfd destination, in this case i must enable via ospf.
Networking, sometimes i love it, mostly i hate it.Its all about the $$$$

Comments

  • joelsfoodjoelsfood Member Posts: 1,027 ■■■■■■□□□□
    I've not used CSRs, but on NXOS you definitely need to enable BFD in the routing process too. This lets BFD know which higher level processes (OSPF, EIGRP, etc) need to be notified when BFD stops receiving replies and tears down an adjacency. Otherwise the routing protocols depend on their own internal timers, which of course can be painfully slow
  • EdTheLadEdTheLad Member Posts: 2,111 ■■■■□□□□□□
    Without BFD on a directly connected fibre, when a physical fibre fails i.e. LOS, i expect the IGP is notified instantly that a port is down. I expect the same behaviour when BFD is running over the bundle, BFD has the quick detection mechanism to find a physical layer issue. Once a port is seen internally as down, the IGP would be informed regardless of BFD.
    btw, unless running in echo mode BFD technically does not receive replies, it's seen as a unidirectional protocol on both sides.

    Just read, to conform to the bfd detection timer you need to tie it to the IGP. So i guess without configuring under the iGP, detection is probably roughly 500ms - 1sec, whereas if you enable under the igp you can get the optimal 50ms detection.
    Networking, sometimes i love it, mostly i hate it.Its all about the $$$$
  • gorebrushgorebrush Member Posts: 2,743 ■■■■■■■□□□
    Yeah you got it - it needs to be tied into the IGP.
  • EdTheLadEdTheLad Member Posts: 2,111 ■■■■□□□□□□
    High end platforms like CRS,NCS have BoB which is BFD over Bundles, this enables BFD on all member ports of a bundle. You do not need to configure BFD on the IGP as any protocol running over BoB will automatically be linked to BFD. BLB which is BFD over logical bundle, will use only a single member and this needs to be specified under the IGP.

    To enable BoB, you specify the destination ip address of the NH interface with the following command "bfd address-family ipv4 destination x.x.x.x" .
    If using just physical interfaces .i.e not bundles, you need to configure the interface with BFD under the IGP.
    Networking, sometimes i love it, mostly i hate it.Its all about the $$$$
Sign In or Register to comment.