This repository contains the code for a simple micro-benchmark of the public key signature algorithms provided by Google's Tink cryptographic primitives library.
For those who don't care to run benchmarks themselves, here are some results we prepared earlier.
Experiment parameters:
- Tink built from commit
4bec791ccf1012d660188f62e086fbbd582722e9, committed 2018-03-30. - 100 round warmup
- 1000 measured rounds, 1000 signing operations per round
- 59 randomized data input bytes per signing operation (similar to eBACS)
- Signing operations are run sequentially on a single thread.
The benchmark was run on an Amazon EC2 t2-micro instance. The results were as follows:
| Algorithm | Median (ns) | Mean (ns) | Std dev. (ns) | Mean ops per second |
|---|---|---|---|---|
| ED25519 | 117860 | 117921 | 1692 | 8480 |
| ECDSA_P256 | 1125607 | 1125284 | 4410 | 889 |
| ECDSA_P384 | 2545451 | 2544765 | 8164 | 393 |
These were derived from the following actual data points:
The red lines in the graph show the mean, the blue ranges show one standard deviation. The Y axis is scaled to include three standard deviations from the mean; some outliers beyond this did exist, but were rare as one might expect.
Benchmarks of ECDSA_P521 have not yet been completed - this key size is significantly slower than the others tested above, and is not a key size that we anticipate using in practice for some time.