Efficient parallel execution of block transactions in blockchain

Anjana, Parwat Singh (2021) Efficient parallel execution of block transactions in blockchain. In: 22nd International Middleware Conference, Middleware 2021, 6 December 2021 through 10 December 2021, Virtual, Online.

Text
2021_Doctoral_Symposium.pdf - Published Version
Restricted to Registered users only
Download (568kB) | Request a copy

Abstract

Miners and validators in current blockchains serially execute block transactions. Such serial execution cannot efficiently utilize modern multi-core resources, consequently hampering system throughput. We propose three approaches to improve blockchain throughput by introducing parallel execution of block transactions. We present a static analysis-based DiPETrans approach that groups the block transactions into independent shards and executes them parallelly in a distributed fashion using a leader-follower method. DiPETrans is empirically evaluated with 5 million actual transactions from the Ethereum blockchain. Since static analysis fails to identify the conflicts precisely, we introduce OptSmart to exploit multi-processing on a multi-core system to improve throughput further. Miners and validators use multiple threads to parallelly execute smart contract transactions (SCTs) in a block. A miner concurrently executes SCTs using optimistic read-write software transactional memory systems (RWSTMs) and saves the non-conflicting SCTs in the concurrent bin and conflicting SCTs in the block graph (BG). Later, validators re-execute SCTs deterministically in parallel to validate the block by using information appended by the concurrent miner. In terms of throughput, optimistic object-based STMs (OSTMs) with higher-level objects are known to outperform RWSTMs. We propose ObjSC approach based on optimistic OSTM, and a counter-based smart multi-threaded validator (SMV) that efficiently detects and rejects malicious blocks proposed by the malicious miners. The simulation result shows that the proposed approaches outperform existing approaches. © 2021 ACM.

[error in script]

IITH Creators: