François Garillot
banner
huitseeker.bsky.social
François Garillot
@huitseeker.bsky.social
Cryptography, decentralized networks, in close proximity to ☕🦀. Ex: {Protocol, Mysten, Dapper} Labs, Meta, …
In 2025, someone arguing you need FHE operating on global shared state to do good private DeFi is probably selling you something. arxiv.org/abs/2103.01193 academic.oup.com/qje/article...
October 24, 2025 at 4:00 PM
The same discretization that hides individual trades also yields fairer pricing and eliminates latency and information-asymmetry rents ... an insight published 5 years earlier than Angeris et al. by authors less than enthusiastic about blockchains.
October 24, 2025 at 4:00 PM
Redesigning the CFMM to batch is challenging but allows both privacy via ZK-proofs _and_ provides better market design: by removing continuous shared state, you make privacy and efficiency coincide.
October 24, 2025 at 4:00 PM
I aimed to convey my enthusiasm for this progress from a practitioner's perspective ... and suggested a few additional directions worth exploring. You can find the recording and slides here: www.garillot.net/talks/2025-...
Modern Multi-proposer consensus implementations
Multi-proposer consensus protocols let multiple validators propose blocks in parallel, breaking the single-leader throughput bottleneck of classic designs. Yet the modern multi-proposer consensus implementation has grown a lot since HotStuff. This workshop explored the implementation details of recent advances – DAG-based approaches like Narwhal and Sui’s Mysticeti – and revealed how implementation details translate to real-world performance gains. We focused on the nitty-gritty: how network communication patterns and data handling affect throughput and latency. New techniques such as Turbine-like block propagation (inspired by Solana’s erasure-coded broadcast) and lazy push gossip broadcasting dramatically cut communication overhead. These optimizations aren’t just theoretical – they enable modern blockchains to process over 100,000 transactions per second with finality in mere milliseconds, redefining what is possible in decentralized systems.
www.garillot.net
June 19, 2025 at 4:00 PM
But in 2025, there's so much more: thanks to the hard work of @alberto_sonnino, @akihidis, Andrey Chursin, Arun Koshy, Mingwei Tian, and others, the Mysticeti implementation in the Sui repo is now modular, structured, and user-friendly, perfect for various projects.
June 19, 2025 at 4:00 PM
There's a clever yet obscure workaround used in Plonky3 (and other places) that allows you to use this method while preserving your spans, thanks to the magic of a drop-in "maybe-parallel" facade.

Find all the details (and a link to the code) in the note:
April 28, 2025 at 4:00 PM
A potential issue with this method, which relies on spans, is that it becomes less useful when using Rayon, used to leverage parallelism in CPU-bound tasks.
April 28, 2025 at 4:00 PM
My latest note explores this delicate balance—because this ain't a simple area; and I think design is about choosing the right trade-offs for the user.

7/7
March 31, 2025 at 4:00 PM
Hybrid architectures might offer a way forward, blending the parallel strengths of BCB with periodic consensus checkpoints or innovative fraud-proof mechanisms.

6/7
March 31, 2025 at 4:00 PM
Another major hurdle: scalability of reads. Without consensus, clients must query multiple validators directly, risking bottlenecks. Is BCB's speed benefit eroded by its read-path complexity?

5/7
March 31, 2025 at 4:00 PM
The challenge: BCB systems inherently limit expressivity and introduce complexity around state contention. Is full parallelism practical, or are consensus-backed approaches like Sui actually better aligned with real-world blockchain usage?

4/7
March 31, 2025 at 4:00 PM
At its core, BCB leverages independent state shards to eliminate global coordination, promising linear complexity over traditional worst-case quadratic Byzantine consensus. Theoretically powerful—but does this parallelism come at a hidden cost?

3/7
March 31, 2025 at 4:00 PM
There's also 1700 sites among the top 1M still using DKIM Keys < 1024 bits, apparently.
dmarcchecker.app/articles/cra...
How We Cracked a 512-Bit DKIM Key for Less Than $8 in the Cloud
dmarcchecker.app
January 10, 2025 at 2:11 PM