Re: [Idr] draft-dickson-idr-second-best-backup-02

Uli Bornhauser wrote:
> Hi Brian,
>
> I worked with your draft and I have a question regarding the modified 
> step c) (section 3.1.):
>
> You wrote that this step is to be repeated without the SECOND BEST 
> paths, if only SECOND_BESTs remain. In other words, this step is not 
> to be repeated, if at least one path *without* SECOND_BEST attribute 
> is remained.
>

Yes, your interpretation is correct.

> However, right after that you wrote that step c) must remove all 
> SECOND_BEST paths. Well, from my point of view, this does not fit to 
> what you wrote above. Can you explain me what you mean at this point?
>

What that statement is meant for, is an explanation for *why* there is 
the need to repeat the step if only SECOND_BESTs remain (or to rephrase 
in the presence of N'th-bests, if no "best" paths remain).
Which is, that there *must* be a path selected, which was a received 
"best" path, in order to proceed to the next step, or to terminate the 
path selection (i.e. because only one path remains, and it is a received 
"best" path).

Perhaps a better way to word it would be, "must have removed all...".

> Another, perhaps more important point, regarding the question (I read 
> in the IEFT-meetings) how many paths (2 or n) are to be sent: You 
> wrote that the approach you presented eliminates persistent 
> oscillations. Provided that I don't misunderstand you, this is not 
> totally right.
> If you consider Appendix B: Oscillation is avoided by your approach as 
> long as you have one path (path b) which may be selected by R4 as best 
> one, while a path (path c) of R4 is required by R1 to reach a stable 
> routing. Generally spoken for this example, you approach works fine, 
> as long as n paths are announced, provided that R4 may select not more 
> than (n-1) paths as best one. If more than (n-1) may be selected by R4 
> as best one, path c is not announced to R1 and oscillation appears.
>
> I attach an extended example based on your ones where the announcement 
> of two signaling paths does not work anymore. It would require that 
> the first, second, and third best is announced. This example can be 
> extended for each static number of paths which are announced by your 
> approach. Thus, to avoid oscillation certainly, a dynamic number of 
> paths must be announceable, for example as described in the walton-draft.
>

Thank you, that is a very helpful example. Would it be alright if I use 
it in the draft?

> Let me know what you think.
I will definitely be modifying the draft, to incorporate N rather than 
just 2 for the additional paths.

I'll be using most of the walton-draft, as it covers the N-path 
codification quite well.

Thanks,

Brian
_______________________________________________
Idr mailing list
Idr@ietf.org
https://www.ietf.org/mailman/listinfo/idr