Performance Benchmarks


I attended the Meetup on Sept 15 2022 and the number that stuck to my head was the 3 orders of magnitude performance number against plain text. I would just like to place this in context. Attached is a result of a recent paper comparing FHE and PHE libraries. That includes Cryptosystem of your famed CTO. My question is - “Where does ZAMA’s FHE fall in this graph?”

Sidorov, Vasily, Ethan Yi Fan Wei, and Wee Keong Ng. “Comprehensive Performance Analysis of Homomorphic Cryptosystems for Practical Data Processing.” arXiv preprint arXiv:2202.02960 (2022).

All schemes will have roughly the same leveled performance (Concrete doesn’t support leveled multiplication yet but its coming soon).

Im not sure however that it makes much sense to benchmark FHE schemes without benchmarking bootstrapping, since this is what the slowdown is due to for the most part. Also since scheme has its own advantages, the best thing would be to compare application benchmarks to have a real view of where things are!