[Sidrops] non-viable proposal (Was: Improving RPKI Efficiency)

[Job Snijders <[e-mail]>]

Hello,

One cannot assume the byte output of a serialized Protobuf message is
stable between different builds, process runs, or languages.
Consequently, Protobuf does not replace X.690/DER. Stable field ordering
obviously is important when hashing messages.
https://protobuf.dev/programming-guides/encoding/#order

Additionally, the paper lacks recognition that the major downside of
RRDP's design isn't its use of XML for encoding, but that clients often
end up fetching data which already had become outdated (e.g. overtaken
by later Delta updates) or fetching data the client already previously
had acquired through another transport protocol. FWIW, the Erik
synchronization protocol does not suffer from either issue.

Furthermore, the proposed approach conflicts with the RPKI CA resource
certificate Key Usage restrictions specified in RFC 6487 section 4.8.4
(also see RFC 5280 section 4.2.1.3). So it seems that despite claims to
the contrary, this proposal is not entirely backwards compatible, also
not as a parallel version of repositories.

An operational concern: it is unclear how an applicant acquires a
signature from the CA without something like PKCS#10.

All in all I don't see anything actionable for the SIDROPS WG.

Regards,

Job


On Tue, Jul 15, 2025 at 09:44:06AM +0200, Niklas Vogel wrote:
> Dear all,
> 
> I hope you are doing well.
> 
> Our team has conducted a research project into RPKI efficiency that fits
> well with the recent mailing list discussions. It would be great to get your
> feedback on the proposed changes to RPKI .
> 
> The paper is uploaded here [1], it will be published at NDSS2026.
> 
> The main efficiency improvements are:
> 
> - Getting rid of EE-Certificates completely
> 
> - Getting rid of signatures on ROAs (and ASPAs etc.) completely. Instead,
> protect them with manifest hashes.
> 
> - Integrate CRL revokedCertificates into the manifest to remove the need for
> a dedicated CRL
> 
> - Group snapshot entries by CA to remove the need for full URIs in each
> entry
> 
> - Use protobuf instead of XML for RRDP objects
> 
> - Use protobuf instead of DER for all objects (except CA certificates)
> 
> With these improvements, full fetches are up to 20x faster (depending on CA
> setup). We tested this on a local version of the full real-world RPKI tree
> and got a 5.8x speed improvement/6.9x smaller average snapshot size.
> 
> Please see the paper for more details on exact design choices and
> evaluations.
> 
> Looking forward to your opinion.
> 
> Cheers,
> 
> Niklas
> 
> 
> [1]: https://arxiv.org/abs/2507.01465
> 
> -- 
> Niklas Vogel
> ATHENE National Research Center for Applied Cybersecurity
> 
> Wiss. Mitarbeiter Goethe-Universität Frankfurt  |  FG Cybersecurity
> Raum  605 |  Robert-Mayer-Straße 10
> 60325 Frankfurt am Main  |  GERMANY
> 
> E-Mail: [e-mail]
> Web: https://cyber.informatik.uni-frankfurt.de
> 
> _______________________________________________
> Sidrops mailing list -- [e-mail]
> To unsubscribe send an email to [e-mail]