This benchmark measures the performance of the
kyber implementation of the BLS signature scheme.
Both threshold signatures and key/signature aggregation can be measured
using the respective commands threshold and aggregate.
All benchmarks are run sequentially on a single CPU core.
Clone this repository, cd into its directory.
The basic version of the benchmark for, respectively, threshold signatures and signature aggregation
with default parameters can be run by executing
go run ./blsbench aggregateand
go run ./blsbench thresholdA quick example of setting benchmarking parameters:
go run ./blsbench aggregate -d 1024 -t 500ms -n 10 -n 1000 will use a random data array of 1024 bytes as the payload to be signed, each operation will be evaluated for 500 milliseconds, and the aggregation operations will be evaluated with 10 and 1000 signatures.
For the complete description of the commands and flags, run
go run ./blsbench -hThe aggregate command always runs benchmarks with signatures on curve G1 and on G2.
For the threshold command, the curve can be selected using the -g option.
Currently, the threshold is hard-coded to be the first lowest integer strictly greater than two thirds of the total number of shares. E.g., if the number of shares is set to any of 4, 5, and 6, the threshold will be 3. This corresponds to a system of n nodes, f of which are potentially (Byzantine-)faulty, with n >= 3f + 1. In such a system, a quorum of 2f + 1 would correspond to the selected threshold for reconstructing the BLS signature.
The VerifyShare operation is currently inefficient - it depends on the threshold selected for the used threshold scheme. This is a property of the implementation and probably can be worked around by pre-computing a point on the curve for each node only once (instead of each time its partial signature is verified) and reusing it for subsequent verifications.