On Tue, Apr 01, 2025 at 01:23:27PM +0200, Tobias Fiebig wrote: > > I don't think we can operate under the assumption that 1/3rd of the > > ASes in the global Internet routing system are single-homed. > > I am referencing information presented by Marco Schmidt (RIPE NCC) at a > recent open house on ASNs here: > > https://ripe89.ripe.net/wp-content/uploads/presentations/88-Personal-ASN.pdf > > I agree that a wider perspective might be useful here, as this might > indeed be collected from RIPE RIS.
I have no idea how RIPE NCC reliably deducts single- vs multi-homed, but I'm very skeptical accurate numbers can be pulled from RIPE RIS. For example, if the AS under investigation is not feeding its full routing table into RIPE IRS and also doesn't have downstream transit customers... then private peering connections would be unlikely to appear in the data-set. I don't think RIPE NCC can conclude whether ASes are single-homed even when those ASes appear behind only a single adjacency in RIS. It might be good to follow up with RIPE NCC on how exactly they think they can reliably observe 'multi-homing'. Please note that "multi-homing" doesn't mean "having multiple transit providers", as I understand the term multi-homing it simply means that an AS has multiple adjacencies to other ASes (for example, 1 upstream and 1 peering partner is multi-homing) > > > - Various anycast use-cases are incompatible with that requirement > > > > Can you elaborate a bit more beyond stating a simple assertion? > > - Using large hosters as 'Anycast' PoPs (frequently done with, e.g., > Vultr) > - The setups for a.root and j.root > - Any Anycast setup with more than one site, as 1930 states: "Without > exception, an AS must have only one routing policy." yeah, the "Without exception, an AS must have only one routing policy." sentence and the practise of using the same ASN in multiple POPs without backbone seem somewhat at odds with each other. > (This may very well be the root-cause of what is happening with the > a.root / j.root ASes) > > Furthermore, prevalent BYOPIP use-cases also see a lot of single homed > ASes. Similarly, I doubt that most ASes still conform to "Without > exception, an AS must have only one routing policy." in general. "prevalent BYOPIP use-cases also see a lot of single homed ASes"? Can you point at these prevalent use-cases? > > > - Technically, IXes are not really compatible with RFC1930 atm > > > either > > > > Can you elaborate a bit more beyond stating a simple assertion? > > An IX is (usually) not originating prefixes. Origination of prefixes > is strictly implied in 1930, as "[...] the exchange of external > routing information alone does not constitute the need for an AS." In my simple reading of RFC1930 IX Route Servers deployments are covered in the Section 5.1 "Multi-homed site" category. Kind regards, Job _______________________________________________ GROW mailing list -- [email protected] To unsubscribe send an email to [email protected]
