Baldur: A Hybrid Blockchain Database with FPGA or GPU Acceleration

Abstract

In the last decade, blockchain technologies have entered the enterprise space in the form of permissioned blockchains, verifiable databases, and hybrid blockchain database systems (HBDB). The core of these blockchain technologies is represented by cryptography operations such as secure hashing typically with SHA-3 algorithms, signature generation, and signature verification, typically done with an Elliptic Curve Digital Signature Algorithm (ECDSA). These cryptography operations incur significant overhead and often they can be done in parallel with other operations such as data fetching and processing. Moreover, these operations can be offloaded to a hardware accelerator. In this paper, starting from the above observations, we present Baldur, an HBDB system in which most of the cryptography operations are offloaded to an accelerator. Specifically, we offload to either a Field Programmable Gate Array (FPGA) or a Graphics Processing Unit (GPU) the Keccak SHA-3 hashing and secp256k1 ECDSA signature verification for the requests received by Baldur from users. We present our preliminary implementation and results for an AMD-Xilinx Varium C1100 FPGA and an Nvidia RTX A6000 GPU. In stand-alone mode, our secp256k1 FPGA implementation achieves a throughput of 12,000 verifications per second, which is 11 × higher compared to the original implementation from AMD-Xilinx. Our GPU implementation achieves a throughput of close to 60,000 verifications per second, which is 25% higher compared to the CPU. When integrated into Baldur, the FPGA and GPU improve the throughput by 12% and 7%, respectively, in the best case. These results are promising but they can be improved, and, hence we plan to further optimize our FPGA and GPU designs and implementations in the future.