Re: [Sidrops] [WGLC] draft-ietf-sidrops-roa-considerations-01 - Ends 10/March/2022

[Geoff Huston <[e-mail]>]

> On 10 Mar 2022, at 11:07 pm, Randy Bush <[e-mail]> wrote:
> 
>>> the CA, in some thought of saving bandwidth or something, packs data
>>> up into AS-based blobs which the RPs are then supposed to unpack and
>>> reconcile.  where's the win that is worth all that work?
>> 
>> “less objects, smaller”
>> 
>> It sounds like an attractive optimisation on paper.
> 
> "Premature optimization is the root of all evil (or at least most of it)
> in programming.”  -- D Knuth
> 
>> I would need benchmark results (with a rir-sized repo with
>> 1-vrp-one-ROA) to see how much it does in practice.
> 
> i think one would want to measure whole system, from CA repo to the
> router
> 
>> My mental model: The RPKI is a directed rooted forest. From a single
>> point of view, for all CAs, you see a consistent snapshot for a CA (or
>> encounter a failed fetch).
> 
> [ s/for all/for each/ ? ]
> 
>> If there are no split views on the repository content, RPs will
>> eventually see the same updates.
> 
> except the CA PPs publish asynchronously, on varied schedules, etc.
> 
> and the RPs fetch asynchronously.
> 
> and the routers ...
> 


How many times do we need to go around and around this space before we realise
that asynchronous demand-pull flooding designs are intrinsically broken?

At least one more time, it seems.

Geoff