Re: [Dcrouting] draft-przygienda-rift-03

[Robert Raszuk <[e-mail]>]

Tony,

> The price of that would be that you can run strict SPF only then and
loose all the non-equal cost forwarding, including true anycast.

1. Did anyone really asked that we should refrain from running "strict SPF"
in DCs ? Contrary I see group of folks keen on running it based on BGP-LS
like feeds :)

2. Is non-ECMP fwd-ing a feature or a bug in DC fabric ? When I described
to customers that with MPLS you can run non ECMP load balancing it was
shock .. but in WAN there are some use cases for it. But in DC fabric - I
doubt it.

3. True anycast is clearly possible with native link state and native
distance vector. Why the new "hybrid" glue of both would not allow it ?
Because it counts on massive aggregation - right ?

> Plus, once you think that through in general sense you basically end up
with host addresses everywhere (since if you use aggregates you'll
blackhole on link failures). If you want to avoid SPF and host routes

Host routes are just leaves so they hang off nodes and do not make SPF any
harder. Besides I am not convinced that building flat underlays with 1M
host routes is the right architecture - even if some folks ask you about
it.

Best,
R.




On Sat, Jan 13, 2018 at 9:05 PM, Tony Przygienda <[e-mail]>
wrote:

>
>
> On Sat, Jan 13, 2018 at 11:21 AM, Robert Raszuk <[e-mail]> wrote:
>
>> Hi Tony,
>>
>> > ​
>> if you're willing to provision things correctly yourself via S-PGP.
>>
>> I am not willing to do that. I would like routing to do it for me
>> auto-magically.
>>
>
> The price of that would be that you can run strict SPF only then and loose
> all the non-equal cost forwarding, including true anycast. Plus, once you
> think that through in general sense you basically end up with host
> addresses everywhere (since if you use aggregates you'll blackhole on link
> failures). If you want to avoid SPF and host routes while still asking for
> all the things you ask, then as a very experienced and highly regarded
> architect here puts it to end such discussions "your problem is overly
> constrained" and there is no protocol/algebra whatsoever that will give you
> that ...
>
>
>> But yes you got the question right. I was asking for shortcuts between
>> last levels of fabric. Not so much between Node 111 & Node 112 - but you
>> said it is optional so ok.
>>
>> Side note: PGP abbrev. for vast majority of people means completely
>> different thing then what you defined it to mean locally in your draft. I
>> highly recommended you rename it in -05 version to PGD (policy guided
>> destination(s) or PGR (policy guided reachability/routing).
>>
>
> Sigh, I warned when the acronym was picked.  Come up with something that
> is a word, the funnier the better, easier to remember for our brains that
> way. Can be 4 letters ;-)  and will get you an Ack mention l;-)
>
>
>>
>> Now requirement to switch off miscabling detection is not acceptable. You
>> are stating that only rift knows how should I cable my fabric ?
>>
>
> No, it doesn't. You asked in your previous email for "red lights" when
> topology is built incorrectly and you have it. Now you don't want it. Both
> states cannot be true @ the same time unless you are entangled ;-)
>
> > Node112 can actually even go haywire
>>
>> That may be not best property of a routing protocol :)
>>
>>
> "haywire" is the wrong word. Rather,  I should have said "it will even
> allow that". Now, why would you want the same src/dst pair sometimes
> shortcut on the horizontal link and sometimes go up to spine I can't
> imagine but you ask for all kind of very non-obvious things you seem to
> like so I'm just pointing the space of possibilities out to you ...
>
>
> -- tony
>