Re: [trill] Asymmetric link costs and D-Tree calculations

[Donald Eastlake <[e-mail]>]

Hi,

As I said I would below, I did file an Errata against RFC 6325.
Although the notice I got saying it had been filled appeared to cc the
TRILL WG mailing list, I don't see that mail in the archive. So I'm
posting a pointer to it here:

http://www.rfc-editor.org/errata_search.php?rfc=6325&eid=3508

Thanks,
Donald
=============================
 Donald E. Eastlake 3rd   [phone] (cell)
 155 Beaver Street, Milford, MA 01757 USA
 [e-mail]


On Fri, Mar 1, 2013 at 2:21 AM, Donald Eastlake <[e-mail]> wrote:
> Hi Olen,
>
> On Thu, Feb 28, 2013 at 1:21 PM, Olen Stokes
> <[e-mail]> wrote:
>> Appendix A.1 of RFC 6325 discusses the fact that “two RBridges may know they
>> are connected but each sees the link as a different speed from how it is
>> seen by the other.”
>
> Yes, and asymmetry can also occur through manual cost configuration of costs.
>
>> We have been looking at this and how it relates to D-Tree calculation as
>> discussed in Section 4.5.1.  In that section, it states, “Each RBridge RBn
>> independently calculates a tree rooted at RBi by performing the SPF
>> (Shortest Path First) calculation with RBi as the root”.  So, it appears
>> that the D-Tree is calculated from the root RBridge outward.
>>
>> The following three paragraphs describe how to determine the potential
>> parents P for a node N.  There does not appear to be any mention of
>> direction so it would seem that all calculations are done in the direction
>> from the root RBi outward.
>
> That was the intent. The tree is calculated through a Dijkstra SPF
> process starting at the Root and working outwards.
>
>> The last paragraph in Section 4.5.1 states, “In other words, the set of
>> potential parents for N, for the tree rooted at R, consists of those that
>> give equally minimal cost paths from N to R and that have distinct IS-IS
>> IDs, based on what is reported in LSPs.”  In this case, costs are discussed
>> as inward from node N towards root RBi.
>
> That text in RFC 6325 is wrong, as far as I know.
>
>> This change in direction of cost calculation could potentially be confusing
>> when there are asymmetric link costs as described in Appendix A.1.  For any
>> parent P found during the SPF calculation, it is possible that the cost from
>> N to
>> P is different than the cost from P to N.  It is also possible that none of
>> the parents found during the SPF calculation are found along any of the
>> “equally minimal cost paths from N to R”.
>
> Yes.
>
> Although I think the asymmetric costs will be rarely encountered, it
> is critical that all RBridges calculate the same tree so the method
> must be unambiguous.
>
>> Is the potential change of direction when calculating costs in the last
>> paragraph unintended?  Would the paragraph have been more clear discussing
>> “equally minimal cost paths from R to N”?
>
> Yes, thanks for spotting this. It should have said "from R to N" or
> something like that. I'll file an Errata against RFC 6325.
>
>> Is there agreement in the presence of asymmetric link costs that it is
>> possible that none of the parents found during the SPF calculation are found
>> along any of the “equally minimal cost paths from N to R” calculated for
>> known unicast traffic?
>
> It seems to me that could happen.
>
> Thanks,
> Donald
> =============================
>  Donald E. Eastlake 3rd   [phone] (cell)
>  155 Beaver Street, Milford, MA 01757 USA
>  [e-mail]
>
>> Cheers,
>> Olen