Description

The repository hosts the implementation of SeerEVM, an advanced transaction execution engine for EVM-compatible blockchains.

SeerEVM incorporates fine-grained branch prediction to fully exploit pre-execution effectiveness. Seer predicts state-related branches using a two-level prediction approach, reducing inconsistent execution paths more efficiently than executing all possible branches. To enable effective reuse of pre-execution results, Seer employs checkpoint-based fast-path execution, enhancing transaction execution for both successful and unsuccessful predictions.

Precondition

Testbeds:

• Amazon EC2 c7i.8xlagre instance (32 vCPUs, 64 GB memory), running Ubuntu 22.04 LTS

Necessary software environment:

• Installation of go (go version >= 1.20)

Remove any previous Go installation by deleting the /usr/local/go folder (if it exists), then extract the archive you just downloaded into /usr/local, creating a fresh Go tree in /usr/local/go:

$ rm -rf /usr/local/go && tar -C /usr/local -xzf go1.22.1.linux-amd64.tar.gz

Add /usr/local/go/bin to the PATH environment variable.

export PATH=$PATH:/usr/local/go/bin

• Install Go-related modules/packages

# Under the current directory `SeerEVM`
$ go mod tidy

Dataset:

Due to the size of the dataset used in our paper exceeding 100 GB, we have trimmed the dataset to facilitate artifact evaluation. We provide a tiny-scale of Ethereum datasets (approximately 370 MB), spanning 1000 blocks (including state data):

• Block height ranging from 14,650,000 ~ 14,651,000, shown in ./ethereumdata_1465_small

Usage

Here, we show how to run SeerEVM to produce several critical experimental results. All the test scripts are presented in the directory ./experiments. Note that due to the different idle state of the machine's resources and the tiny scale of Ethereum datasets—which results in minimal disk I/O operations—it is normal for the reproducible results to have deviations with those reported in the paper.

1.1 Prediction accuracy

Enter the directory ./experiments and run the script prediction_height.sh:

$ cd experiments
$ ./prediction_height.sh

This script run would output the prediction accuracy across 1,000 blocks shown in prediction_height.txt. This experimental result presents the prediction accuracy of complex contract transactions (C-TXs) and normal contract transactions (N-TXs) at every 200-block offset.

1.2 Pre-execution latency

In the current directory, run the script pre_execution_lat_large.sh:

$ ./pre_execution_lat_large.sh

This script run would output the pre-execution latency under different number of transactions shown in preExecution_large_ratio.txt. For each transaction scale, three different disorder rates (0, 0.4, 0.8) are used. Under each disorder rate, the pre-execution latency of both C-TXs and N-TXs is presented, along with the total latency of pre-executing all transactions.

1.3 Speedup over single transaction execution

In the current directory, run the script speedup_tx.sh:

$ ./speedup_tx.sh

This script run would output the speedup distribution across smart contract transactions shown in speedup_perTx_full.txt. This experimental result presents the average speedups on C-TXs and N-TXs, as well as the average speedup for all transactions.

1.4 Concurrent execution performance

In the current directory, run the script con_execution_abort.sh:

$ ./con_execution_abort.sh

This script run would output the abort rate under varying number of concurrent transactions shown in concurrent_abort.txt. This experimental result presents the average abort rates for C-TXs and N-TXs.

Then, run the script con_execution_speedup.sh:

$ ./con_execution_speedup.sh

This script run would output the speedup over serial execution by using different threads shown in concurrent_speedup.txt. This experimental result presents the latency and throughput of SeerEVM's concurrent execution, alongside those of the serial execution for comparison. Note that, compared to the full-node storage size simulated in the paper, the dataset used here is relatively small. As a result, the reduction in disk I/O overhead achieved by SeerEVM's pre-execution over serial execution is less significant, leading to a lower speedup than reported in the paper.

Citation

If you are interested in our work and intend to use SeerEVM in your research, please consider citing our paper. Below is the BibTex format for citation:

@article{zhang2024seer,
  title={Seer: Accelerating Blockchain Transaction Execution by Fine-Grained Branch Prediction},
  author={Zhang, Shijie and Cheng, Ru and Liu, Xinpeng and Xiao, Jiang and Jin, Hai and Li, Bo},
  journal={Proceedings of the VLDB Endowment},
  volume={18},
  number={3},
  pages={822--835},
  year={2024},
  publisher={VLDB Endowment}
}