3 ms·
It could be, but most practical FHE are ridiculously slow and not usable past trivial data sizes. It would be cool if they published some numbers for cupcake.
by viraptor 5y ago
It could be, but most practical FHE are ridiculously slow and not usable past trivial data sizes. It would be cool if they published some numbers for cupcake.
Looking at an older paper: https://eprint.iacr.org/2011/405.pdf https://eprint.iacr.org/2011/405.pdf "Homomorphic addition is essentially instantaneous (i.e., takes less than 1 ms), whereas homomorphic multiplication takes about 41 ms".
My run (with decyption disabled because it didn't compile):
test encrypt_pk ... bench: 619,097 ns/iter (+/- 38,166)
test encrypt_sk ... bench: 540,205 ns/iter (+/- 40,429)
test encrypt_zero_pk ... bench: 576,200 ns/iter (+/- 39,531)
test homomorphic_addition ... bench: 21,842 ns/iter (+/- 1,088)
test rerandomize ... bench: 591,631 ns/iter (+/- 23,246)
So 21us for addition - that's getting into a reasonable territory, but may be close to the result from the paper. Multiplication doesn't seem to be supported yet.