OSPF Fast Convergence (RFC6976)?
Hi all, I came across ``Framework for Loop-Free Convergence – Using the Ordered Forwarding Information Base (oFIB) Approach'' [0] which may be a phantastic remedy to a problem that occurs from time to time in a network I maintain: Even restarting a single OSPF process (on a group of about 20 routers in that area) causes trouble in forwarding/routing of 60.000+ TCP streams that traverse that area. (EIGRP is said to be immune against that because of explicit termination of the diffusing computations.) AFAICS it's not implemented yet in any free routing suite – are there any plans of doing so (in FRR)? Best regards, Timo [0] – https://tools.ietf.org/html/rfc6976
Timo - We have not implemented this yet. We would welcome someone to do this work. I have added this to the feature page https://github.com/FRRouting/frr/wiki/Feature-Requests donald On Tue, Aug 14, 2018 at 7:56 AM, Timo Schöler <timo@kroenchenstadt.de> wrote:
Hi all,
I came across
``Framework for Loop-Free Convergence – Using the Ordered Forwarding Information Base (oFIB) Approach'' [0]
which may be a phantastic remedy to a problem that occurs from time to time in a network I maintain: Even restarting a single OSPF process (on a group of about 20 routers in that area) causes trouble in forwarding/routing of 60.000+ TCP streams that traverse that area.
(EIGRP is said to be immune against that because of explicit termination of the diffusing computations.)
AFAICS it's not implemented yet in any free routing suite – are there any plans of doing so (in FRR)?
Best regards,
Timo
[0] – https://tools.ietf.org/html/rfc6976
_______________________________________________ frog mailing list frog@lists.frrouting.org https://lists.frrouting.org/listinfo/frog
On 14 August 2018 at 12:56, Timo Schöler <timo@kroenchenstadt.de> wrote:
Hi all,
I came across
``Framework for Loop-Free Convergence – Using the Ordered Forwarding Information Base (oFIB) Approach'' [0]
which may be a phantastic remedy to a problem that occurs from time to time in a network I maintain: Even restarting a single OSPF process (on a group of about 20 routers in that area) causes trouble in forwarding/routing of 60.000+ TCP streams that traverse that area.
(EIGRP is said to be immune against that because of explicit termination of the diffusing computations.)
AFAICS it's not implemented yet in any free routing suite – are there any plans of doing so (in FRR)?
Best regards,
Timo
Hi Timo, Have you seen this darft? https://tools.ietf.org/html/draft-ietf-rtgwg-backoff-algo-10 It discusses a simpler method to achieve similar results. RFC6979 isn't implemented anywhere as far as I know, even by Cisco for example who co-wrote that RFC. I think in reality it wasn't the best approach. If you use a heirarchical FIB then after detecting a failure the traffic loss from reconvergence alone can be as low as a few milliseconds. Cheers, James.
On 08/14/2018 02:34 PM, James Bensley wrote:
On 14 August 2018 at 12:56, Timo Schöler <timo@kroenchenstadt.de> wrote:
Hi all,
I came across
``Framework for Loop-Free Convergence – Using the Ordered Forwarding Information Base (oFIB) Approach'' [0]
which may be a phantastic remedy to a problem that occurs from time to time in a network I maintain: Even restarting a single OSPF process (on a group of about 20 routers in that area) causes trouble in forwarding/routing of 60.000+ TCP streams that traverse that area.
(EIGRP is said to be immune against that because of explicit termination of the diffusing computations.)
AFAICS it's not implemented yet in any free routing suite – are there any plans of doing so (in FRR)?
Best regards,
Timo
Hi Timo,
Have you seen this darft?
https://tools.ietf.org/html/draft-ietf-rtgwg-backoff-algo-10
Hi James, no, I hadn't seen it yet. Sounds like a more reasonable technique than from the RFC I referenced. @Donald: Maybe you want to modify it appropriately on https://github.com/FRRouting/frr/wiki/Feature-Requests ?
It discusses a simpler method to achieve similar results. RFC6979 isn't implemented anywhere as far as I know, even by Cisco for example who co-wrote that RFC. I think in reality it wasn't the best approach.
If you use a heirarchical FIB then after detecting a failure the traffic loss from reconvergence alone can be as low as a few milliseconds.
Cheers, James. Best,
Timo
participants (3)
-
Donald Sharp -
James Bensley -
Timo Schöler