Parallelized multi-backend KZG library for Ethereum Data Sharding (aka Proto-Danksharding, EIP-4844)
The goal is to create a parallelized KZG library for Ethereum Data Sharding (aka Proto-Danksharding, EIP-4844) that supports multiple ECC (Elliptic-curve cryptography) backend libraries.
Support for multiple backend ECC libraries is implemented via Traits. Such an approach allows to easy change backend ECC libraries as all the crates shared the same interface (see benchmarks and tests). The current state of supported backend ECC libraries:
| Backend ECC | FFT/DAS | EIP-4844 (non-parallel) | EIP-4844 (parallel) | c-kzg-4844 drop-in replacement | GPU acceleration |
|---|---|---|---|---|---|
| blst | ✔️ | ✔️ | ✔️ | ✔️ | ✔️ via sppark |
| constantine | ✔️ | ✔️ | ✔️ | ✔️ | ❌ |
| mcl | ✔️ | ✔️ | ✔️ | ❌ | ❌ |
| arkworks | ✔️ | ✔️ | ✔️ | ✔️ | ❌ |
| arkworks3 | ✔️ | ✔️ | ✔️ | ✔️ | ✔️ via sppark and wlc_msm |
| zkcrypto | ✔️ | ✔️ | ✔️ | ✔️ | ❌ |
We aim to expose an identical C interface compared to c-kzg-4844 so that rust-kzg could work as a drop-in replacement for c-kzg-4844. If you already use c-kzg-4844 bindings you can try faster paralellized rust-kzg without any changes to your code-base by simply replacing the binary. Instructions for C#, Java, Nodejs, Python, Rust bindings are available here.
The best place to look for examples is tests directory.
Currently, the ECC backend is set by pointing Cargo to the corresponding crate:
[dependencies]
kzg = { git = "https://github.com/sifraitech/rust-kzg.git", package = "rust-kzg-blst" }
kzg_traits = { git = "https://github.com/sifraitech/rust-kzg.git", package = "kzg" }
Benchmarks run on every Github build. However, it's best to run them on a dedicated machine. Tautvydas rendered nice charts for results he got on cloud servers:
Click to expand (blobs count 32 to 1)
GPU-accelerated multi-scalar multiplication is available for arkworks3 and blst backends.
The project is mainly developed by a group of students at the Blockchain Technologies course led by Saulius Grigaitis. The project is heavily based on the go-kzg, c-kzg, and other libraries.