Stm reduce transmute #675
Stm reduce transmute #675
Conversation
curiecrypt
force-pushed
the
stm-reduce-transmute
branch
from
1524a6c
to
3c00e04
Compare
curiecrypt
temporarily deployed
to
testing-preview
GitHub Actions
Inactive
| let ser_sk = sk.0.serialize().as_ptr(); | ||
| let mut sk_scalar = blst_scalar::default(); | ||
| blst_scalar_from_bendian(&mut sk_scalar, ser_sk); | ||
| let mut out = blst_p1::default(); | ||
| blst_sk_to_pk_in_g1(&mut out, sk_scalar); | ||
| blst_sk_to_pk_in_g1(&mut out, &sk_scalar); |
There was a problem hiding this comment.
In general I would avoid doing serialisation/deserialisation, and use it as a last resort. So if for the other usages of transmute we can use different strategies, it'll be best.
If we were to continue with this technique (of serialisation and deserialisation), I would be careful with secret material. i.e. I would overwrite ser_sk and sk_scalar with zeroes and blst_scalar::default respectively.
curiecrypt
temporarily deployed
to
testing-preview
GitHub Actions
Inactive
curiecrypt
temporarily deployed
to
testing-preview
GitHub Actions
Inactive
curiecrypt
temporarily deployed
to
testing-preview
GitHub Actions
Inactive
|
As mentioned by @curiecrypt, the regression is considerable, and we won't proceed on that path. Doing serialisation/deserialisation, while being safer, has an important hit on performance. In order to minimise the risks of transmute, we will create helper function (to facilitate auditing), and fix to a particular version of |
Content
This PR aims to reduce or completely remove the use of
transmuteinmulti_sig.rs.Pre-submit checklist
Comments
The following are the benchmark results giving the regression between the versions with
transmuteandserde. The performance has regressed drastically. So, we should consider another way to prevent the possible problems of usingtransmutewhich is also efficient. As we discussed, a helper structure for unsafe codes could be better @iquerejeta.STM/Blake2b/Key registration/k: 25, m: 150, nr_parties: 300 time: [502.24 ms 502.95 ms 503.73 ms] change: [+26.772% +27.320% +27.940%] (p = 0.00 < 0.05) Performance has regressed. STM/Blake2b/Play all lotteries/k: 25, m: 150, nr_parties: 300 time: [908.94 µs 910.58 µs 912.56 µs] change: [+31.221% +31.619% +32.056%] (p = 0.00 < 0.05) Performance has regressed. STM/Blake2b/Aggregation/k: 25, m: 150, nr_parties: 300 time: [24.401 ms 24.622 ms 24.895 ms] change: [+26.608% +27.610% +29.137%] (p = 0.00 < 0.05) Performance has regressed. STM/Blake2b/Key registration/k: 250, m: 1523, nr_parties: 2000 time: [3.3481 s 3.3779 s 3.4273 s] change: [+23.353% +27.009% +30.851%] (p = 0.00 < 0.05) Performance has regressed. STM/Blake2b/Play all lotteries/k: 250, m: 1523, nr_parties: 2000 time: [7.5255 ms 7.5313 ms 7.5375 ms] change: [+30.692% +30.971% +31.228%] (p = 0.00 < 0.05) Performance has regressed. STM/Blake2b/Aggregation/k: 250, m: 1523, nr_parties: 2000 time: [238.58 ms 238.82 ms 239.12 ms] change: [+24.642% +24.856% +25.077%] (p = 0.00 < 0.05) Performance has regressed. STM/Blake2b/Batch Verification/k: 25, m: 150, nr_parties: 300/batch size: 1 time: [2.9301 ms 2.9354 ms 2.9427 ms] change: [+30.118% +30.369% +30.663%] (p = 0.00 < 0.05) Performance has regressed. STM/Blake2b/Batch Verification/k: 25, m: 150, nr_parties: 300/batch size: 10 time: [23.629 ms 23.650 ms 23.674 ms] change: [+31.475% +31.664% +31.845%] (p = 0.00 < 0.05) Performance has regressed. STM/Blake2b/Batch Verification/k: 25, m: 150, nr_parties: 300/batch size: 20 time: [46.031 ms 46.192 ms 46.431 ms] change: [+32.254% +32.681% +33.157%] (p = 0.00 < 0.05) Performance has regressed. STM/Blake2b/Batch Verification/k: 25, m: 150, nr_parties: 300/batch size: 100 time: [230.16 ms 230.38 ms 230.62 ms] change: [+31.979% +32.165% +32.361%] (p = 0.00 < 0.05) Performance has regressed. STM/Blake2b/Batch Verification/k: 250, m: 1523, nr_parties: 2000/batch size: 1 time: [4.3937 ms 4.4235 ms 4.4583 ms] change: [+18.011% +19.305% +20.655%] (p = 0.00 < 0.05) Performance has regressed. STM/Blake2b/Batch Verification/k: 250, m: 1523, nr_parties: 2000/batch size: 10 time: [45.043 ms 45.313 ms 45.601 ms] change: [+22.106% +23.430% +24.636%] (p = 0.00 < 0.05) Performance has regressed. STM/Blake2b/Batch Verification/k: 250, m: 1523, nr_parties: 2000/batch size: 20 time: [97.238 ms 100.43 ms 104.30 ms] change: [+44.286% +49.695% +54.132%] (p = 0.00 < 0.05) Performance has regressed. STM/Blake2b/Batch Verification/k: 250, m: 1523, nr_parties: 2000/batch size: 100 time: [396.87 ms 397.68 ms 398.52 ms] change: [+34.851% +35.222% +35.516%] (p = 0.00 < 0.05) Performance has regressed.Issue(s)
Closes #532