This repository contains benchmarks and exploratory code for Union Core V2 on RISC Zero. The objective is to identify the best architecture to efficiently aggregate multiple types of proofs, and potentially supercede Galois.

Union Core V2 aggregates lightclient proofs for an arbitrary amount of chains in parallel, to form the PreState. Once generated, the PreState is transformed into the final State by applying different smart contracts (transformers) over the order flow, performing such operations as netting, CoW, or prefilling.

flowchart TB

A[CometBLS] -->|Galois| I(SNARK)
A[CometBLS] -->|RISC Zero| B(PreState)
I[SNARK] -->|RISC Zero| B(PreState)
D[Sync-Committee] -->|RISC Zero| B(PreState)
E[Tendermint] -->|RISC Zero| B(PreState)
F[Aptos] -->|RISC Zero| B(PreState)
G[PoW] -->|RISC Zero| B(PreState)
B(PreState) --> |RISC Zero| H(State)

For the initial exploration, we are interested in evaluation of this branch:

flowchart TB

A[CometBLS] -->|Galois| I(SNARK)
A[CometBLS] -->|RISC Zero| B(PreState)
I[SNARK] -->|RISC Zero| B(PreState)

In particular, if it is more efficient to aggregate a proof generated by Galois, or to prove the CometBLS lightclient using RISC Zero. In both cases, RISC Zero is used to construct the final proof.

We are interested in the total latency from the first proof until PreState, measured in milliseconds. That means for the CometBLS-RISC Zero approach, we benchmark STARK -> SNARK, while for CometBLS-Galois, we benchmark SNARK -> STARK -> SNARK.

Ensure you have the following installed on your system:

• Rust
• RiscZero toolchain

curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
export PATH=$PATH:$HOME/.cargo/bin
rustup toolchain install nightly
rustup default nightly

curl -L https://risczero.com/install | bash
export PATH=$PATH:$HOME/.risc0/bin
rzup install

The test run various test scenarios. The statistics of running the test is written to ./host/out.csv.

RISC0_DEV_MODE=1 cargo test -p host -- --test-threads=1 --nocapture

To run the tests generating the proof on the local machine, use the following command:

RISC0_DEV_MODE=0 cargo test -p host -- --test-threads=1 --nocapture

test-threads=1: because we run one test at the time, so we can measure performance of each test.

To run the tests while integrating with the Bonsai API, use the following command:

RISC0_DEV_MODE=0 SNARK_WRAPPING=0 BONSAI_API_URL=https://api.bonsai.xyz BONSAI_API_KEY=<your-api-key> cargo test -p host -- --test-threads=1 --nocapture

test-threads=1: because we run one test at the time, so don't run into api usage limitations.

Replace <your-api-key> with your actual Bonsai API key.

To run the tests while integrating with the Bonsai API, use the following command:

RISC0_DEV_MODE=0 SNARK_WRAPPING=1 BONSAI_API_URL=https://api.bonsai.xyz BONSAI_API_KEY=<your-api-key> cargo test -p host -- --test-threads=1 --nocapture

test-threads=1: because we run one test at the time, so don't run into api usage limitations.

Replace <your-api-key> with your actual Bonsai API key.

• System: MacBook Pro (16-inch, Nov 2023) - Apple M3 Max - 128GB - macOS Sonoma 14.6.1
• r0.1.79.0-2-risc0-rust-aarch64-apple-darwin (rustc 1.79.0-dev (22b036206 2024-08-21))

• System: Pop!_OS 22.04 LTS - Intel(R) Core(TM) i9-14900K - 96GB RAM - NVIDIA RTX 4080 - Samsung SSD 990 PRO 2TB rustc 1.79.0 (129f3b996 2024-06-10)

• System: MacBook Pro (16-inch, Nov 2023) - Apple M3 Max - 128GB - macOS Sonoma 14.6.1
• r0.1.79.0-2-risc0-rust-aarch64-apple-darwin (rustc 1.79.0-dev (22b036206 2024-08-21))

• System: Pop!_OS 22.04 LTS - Intel(R) Core(TM) i9-14900K - 96GB RAM - NVIDIA RTX 4080 - Samsung SSD 990 PRO 2TB rustc 1.79.0 (129f3b996 2024-06-10)

• System: MacBook Pro (16-inch, Nov 2023) - Apple M3 Max - 128GB - macOS Sonoma 14.6.1
• r0.1.79.0-2-risc0-rust-aarch64-apple-darwin (rustc 1.79.0-dev (22b036206 2024-08-21))

• System: Pop!_OS 22.04 LTS - Intel(R) Core(TM) i9-14900K - 96GB RAM - NVIDIA RTX 4080 - Samsung SSD 990 PRO 2TB rustc 1.79.0 (129f3b996 2024-06-10)

Below the profiler report of the test_invalid_block_969002 test (all tests have similar results). The raw data can be downloaded here.

Risc Zero offers a straightforward API for creating Zero-Knowledge (ZK) circuits. However, it currently falls short of meeting our performance requirements. Generating a STARK proof and wrapping it in a SNARK proof takes between 1 to 2.5 minutes, whereas our current solution generates the SNARK proof in just 8 seconds.

We assume that introducing elliptic curve accelerators for BN254 (namely add/mul/pairing) would drastically lower proving time. We use the substrate-bn library. This is supported by the profiler report above.

Additional challenges include:

• The toolchain is locked to a specific Rust version, which may conflict with other components of our stack.
• The toolchain requires manual installation, which is incompatible with our current setup using Nix Flakes for development environment management.