Loading...
Accelerating Blockchain Transaction Verification With Parallel Computing
Zhou, Huangjin
Zhou, Huangjin
Files
35.232-2024.07a Huangjin Zhou.pdf
Adobe PDF, 1.14 MB
- Embargoed until 2026-05-18
Description
A Master of Science thesis in Computer Engineering by Huangjin Zhou entitled, “Accelerating Blockchain Transaction Verification with Parallel Computing”, submitted in May 2024. Thesis advisor is Dr. Gerassimos Barlas. Soft copy is available (Thesis, Completion Certificate, Approval Signatures, and AUS Archives Consent Form).
Abstract
Blockchain technology has emerged as a groundbreaking distributed ledger system, within this decentralized network, all historical transactions are recorded in blocks and synchronized across the entire network through block propagation among nodes. To maintain the network's security and data integrity, nodes undergo rigorous verification processes upon receiving new blocks. The most computationally demanding aspect of this process is the verification of each transaction's digital signature within the block. Despite the efficiency of elliptic curve digital signature algorithms used by prominent blockchain platforms like Bitcoin (e.g., Secp256k1), the process often underutilizes the parallel computing capabilities of contemporary GPUs and multi-core CPUs. In this study, we extract and store hash values, digital signatures, and public keys from massive volumes of Bitcoin block data. These signatures are then imported and verified using multi-core CPUs, GPUs, and clusters. Our experiments reveal that on a single-machine basis, including multiple-core CPU computing, CPU+GPU heterogenous computing and pure GPU computing, the speedup of multi-core parallel computing compared to single-core CPU performance can vary between 5 to 50 times. Given the distinct architectures of GPUs and CPUs, the speedup of GPUs is not inherently greater than that of CPUs. Nevertheless, the hybrid scheme, by fully leveraging the computational resources of a single machine, achieves a higher speedup compared to the CPU and GPU schemes individually. By transitioning to distributed computing and forming a cluster of five machines, we achieve a speedup up to 12 times higher than a single machine, incurring only up to 16% in communication and scheduling overhead.